Both the Smoothing combo (this PR) and the existing filter-family
combo had custom inline stylesheets that defined `::drop-down`
geometry but no `::down-arrow` image, so Qt couldn't render the
indicator chevron.

Switch both to the shared `AetherSDR::applyComboStyle()` helper
(matches Phone/CW applet, USB Cables, etc.) — paints a small
triangle pixmap as the down-arrow.  Visual style is otherwise
near-identical (the helper uses #1a2a3a vs the EQ editor's
slightly darker #0e1b28; difference is imperceptible against the
EQ editor's background).

Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>

Draws a 1 px dashed yellow line at 3 kHz on every EQ canvas — marks
the upper edge of the standard SSB voice passband so operators can
see how much of their EQ shape lands inside vs outside it.

Color is QColor(220, 200, 80, 110) — yellow at ~43% alpha so it
reads as a guide, not a foreground element.  Placed in the draw
order between the 0 dB reference line and the EQ analyzer/curves
so it sits behind everything interactive.

Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>

Replaces the static 3 kHz reference line with two dashed yellow
verticals that mirror the radio's current TX filter cutoffs (the
Low Cut and High Cut values shown in the Phone applet).  Operators
now see exactly where their EQ shape lands relative to what's
actually passed to the radio.

Wiring:
- ClientEqCurveWidget gains setFilterCutoffs(lowHz, highHz); 0 for
  either edge skips that guide.
- ClientEqApplet (docked, Tx-bound) and ClientEqEditor (floating)
  expose pass-through setTxFilterCutoffs(); both no-op on RX path.
- MainWindow connects TransmitModel::phoneStateChanged to a lambda
  that pushes current cutoffs to whichever views exist.  Initial
  push runs once at MainWindow ctor + once when the lazy editor is
  first created.
- The editor caches its last-known TX cutoffs so swapping RX → TX
  via showForPath restores the guides without waiting for the next
  phoneStateChanged event.

The RX EQ canvas never sees cutoffs (path-gated at applet/editor
level) so its appearance is unchanged.

Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>

Previous commit subscribed the EQ-canvas cutoff-guide push to
TransmitModel::phoneStateChanged.  That signal fires on every TX
status update touching VOX, CW, mic boost, dexp, AM carrier,
swap_paddles, iambic, sidetone toggles, etc. — many of which run
repeatedly during normal TX.  Three other listeners already do real
work on every fire (iambic keyer start/stop, sidetone gen
setEnabled/setVolume/setPitch/setPan, PhoneCwApplet UI sync); piling
canvas repaints onto that bus appears to have starved the TX audio
path enough that the user reported audible cutouts every ~minute.

Adds a dedicated TransmitModel::txFilterCutoffChanged(lo, hi) signal
that fires *only* when the lo / hi keys arrive AND their values
actually change.  Switches the cutoff-guide listener to it.  Cutoffs
change when the user drags the Phone applet sliders — basically never
during a TX session — so the new signal is silent except when the
guide actually needs to move.

Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>

Toolbar setup reads ClientEqSmoothingFraction from AppSettings and
calls m_canvas->setSmoothingOctaveFraction(savedFraction), but the
toolbar is built ~140 lines before the canvas is constructed.  The
guarded `if (m_canvas)` silently skipped — combo correctly reflected
the saved value but the canvas kept its default (96 = off), so the
user saw "1/6" in the combo with a raw FFT trace.

Cache the saved fraction on a member during toolbar setup; apply it
to the canvas immediately after construction.  Persistence now works
end-to-end: combo + canvas both reflect the saved value on launch.

Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>

Mirrors the TX-side cutoff-guide work onto the RX EQ canvas.  Source
of truth is the *active* RX slice's filter_lo / filter_hi values,
converted from protocol offsets to audio-frequency domain so they
land correctly on the EQ canvas's log-Hz axis:

  USB (lo=0,    hi=3000):   audio 0–3000 Hz
  LSB (lo=-3000, hi=0):     audio 0–3000 Hz (reflected)
  AM  (lo=-3000, hi=3000):  audio 0–3000 Hz (carrier-centered)
  CW  (lo=500,   hi=700):   audio 500–700 Hz (around pitch)

Formula:
  audio_low  = same-sign-or-zero ? min(|lo|, |hi|) : 0
  audio_high = max(|lo|, |hi|)

Wiring:
- ClientEqApplet / ClientEqEditor gain setRxFilterCutoffs(lo, hi)
  (Rx-only, parallel to setTxFilterCutoffs).  Editor caches values
  so RX↔TX path swaps restore the right guides.
- MainWindow::pushRxFilterCutoffsToEq() reads the active slice and
  pushes converted values to the docked applet + floating editor.
- Called on active-slice swap (in setActiveSlice) and on every
  slice's filterChanged signal — connected per slice in the
  sliceAdded handler so click-to-tune-then-drag-filter updates live.

Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>

Adds a 14-px tall colored strip across the bottom of the EQ canvas
(both RX and TX views) marking the common audio-domain modulation
regions:

   0 –   99 Hz : E-SSB    (CW General color, blend 0.40)
 100 – 3 kHz   : SSB      (Phone General color, blend 0.40)
 3k  – 4 kHz   : E-SSB    (CW General color, blend 0.40)
 4k  – 6 kHz   : Voodoo   (CW Extra color, blend 0.20)
 6k  – 20 kHz  : AM / FM  (Data color, blend 0.40)

Colors taken from resources/bandplans/arrl-us.json (CW=#3060ff,
Phone=#ff8000, Data=#c03030); blend factors mirror the
panadapter's band-plan license-class shading so the strip reads
visually as a continuation of the panadapter band.

Reserves kAudioBandStripH = 14 px from the bottom of the drawing
rect — dbToY / yToDb and the FFT analyzer's local h are clipped
above the strip, freq labels move up to sit above it.  Curve and
analyzer content can no longer paint into the bottom 14 px; the
strip is opaque and drawn last so it covers anything that does.

Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>

The 0–99 Hz E-SSB segment's left edge maps to off-screen-left under
freqToX's log scale (which floor-clamps below 20 Hz to -inf, then
linearly extrapolates).  Centering the label across the unclipped
rect put its midpoint hundreds of pixels left of the canvas, so it
never rendered.  Compute label rect from clamped (visLeft, visRight)
so the text lands inside the visible band.

The canvas's log-frequency axis starts at kMinHz = 20 Hz; rendering
the first E-SSB segment from 0 Hz pushed its left edge off-screen
and required a visible-portion clip hack to position its label.
Since the canvas can't display anything below 20 Hz, just bound the
segment data to the canvas range.  Reverts the clip workaround.

Drops the 4-6 kHz 'Voodoo' segment and extends the upper E-SSB band
to cover the full 3-6 kHz range.  Now four segments instead of five.

Dashed yellow lines on the floating editor canvas are now interactive.
Hover shows a horizontal-resize cursor; click+drag horizontally
adjusts the matching filter edge live.  Band-handle hits still take
priority over cutoff-line hits when they overlap.

Wiring:
- ClientEqEditorCanvas adds hitTestCutoffEdge() with ±5 px tolerance,
  excluding the bottom band-plan strip.  Drag emits cutoffsDragged(lo, hi)
  in audio-domain Hz.
- ClientEqEditor forwards the canvas signal as a path-tagged
  cutoffsDragRequested(path, lo, hi) so MainWindow knows which model
  to write to.
- MainWindow dispatches:
    TX → TransmitModel::setTxFilterLow / setTxFilterHigh (direct
         audio-domain).
    RX → SliceModel::setFilterWidth on the active slice with
         mode-aware conversion:
           USB / DIGU / FDV / CW / RTTY: pass-through
           LSB / DIGL:                   sign-flipped & swapped
           AM / SAM / FM / DSB / NFM / DFM:
                 symmetric around carrier — only audio_high
                 controls width, low is mirrored to -audio_high.

Drag clamps: lo ∈ [20, hi-50], hi ∈ [lo+50, 10000].  50 Hz minimum
span matches FlexLib's filter-width floor (Slice.cs).  Canvas
updates locally each frame so the line tracks the cursor without
waiting for the radio's status echo; the radio's echo lands on
the next phoneStateChanged / filterChanged event.

Docked applet remains read-only — same pattern as the existing
band-handle drag (mouse handlers only on ClientEqEditorCanvas).

Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>

Step buttons (and scroll wheel) used to add/subtract 50 from the
current value, so a non-aligned value like 87 Hz would step to 137
or 37 instead of the more useful 100 or 50.

Now the up arrow snaps to the next-higher multiple of 50 and the
down arrow snaps to the next-lower multiple of 50, regardless of
where the current value lands.  Existing min/max clamps preserved.

  current     ▲ goes to     ▼ goes to
  ─────────  ───────────   ───────────
  87 Hz       100 Hz         50 Hz
  100 Hz      150 Hz          50 Hz
  149 Hz      150 Hz         100 Hz
  150 Hz      200 Hz         100 Hz

Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>

applyClientEqTx{Int16,Float32} early-returned when m_clientEqTx
was null or disabled, which also skipped the tapClientEqTx*() call
that feeds the editor's TX FFT analyzer.  Result: when the user
bypassed the TX EQ in the CHAIN widget, the analyzer in the editor
showed no signal — even though mic audio was flowing to the radio
just fine.  Same with anything else that relies on the post-EQ
sample stream.

Restructure both functions so the tap always fires.  When the EQ
is disabled, the tap captures the pass-through samples (identical
to post-EQ since no processing happened).  When enabled, it
captures the post-EQ output as before.  Either way, the analyzer
sees live audio whenever the radio sees it.

Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>