fix(amulegui): keep stats + Kad graphs populated across tab switches - #135

Fixes #134 (stats + Kad parts).

Two visible regressions on amulegui's Statistics tab and the Network → Kad sub-tab, both rooted in the same gap: COScopeCtrl::PlotHistory was an empty stub under CLIENT_GUI because the monolithic implementation calls theApp->m_statistics->GetHistory(...) which had no analogue on the remote-GUI side. Two symptoms:

1. Background starvation — CamuleRemoteGuiApp::OnPollTimer only fired statgraphs->DoRequery() and stattree->DoRequery() when the Statistics dialog was visible. CStatGraphRem::HandlePacket also feeds m_kademliawnd->UpdateGraph(), so leaving Statistics hidden also starved the Kad node-count graph. Switch off Stats → both graphs froze.
2. Whole-graph resets — auto-rescale paths (SetRanges on Kad node count crossing a 50-step ceiling; SetRange on the Stats connection-scope nScale change) call RecreateGraph, which clears the bitmap and then calls PlotHistory(width, false, ...). Under CLIENT_GUI that was a no-op, so the bitmap stayed cleared and the entire line vanished.

What this PR does

Commit 1 — polling-gate fix. Both stats requests now run every step 2 cycle regardless of the active tab. Cost is unchanged in practice: the graph request is a delta (m_lastTimestamp lower bound; daemon returns only newer points or EC_OP_FAILED "No points for graph.") and the tree request is already throttled by thePrefs::GetStatsInterval().

Commit 2 — share PlotHistory with monolithic via a client-side history ring. Rather than duplicate the monolithic machinery, CPreciseRateCounter, CStatistics::GetHistory and CStatistics::ComputeAverages are lifted out of the #ifndef CLIENT_GUI gate alongside the already-shared HR struct. The CLIENT_GUI CStatistics now also carries listHR, the three m_graphRunningAvg* counters and hrInit. A new CLIENT_GUI-only producer AddHistoryRecord(const HR&) is called once per decoded point from CStatGraphRem::HandlePacket. The ring is capped at kHistoryCap = 1800 (~30 min @ 1 Hz, ≈150 KB across all four graphs). With listHR populated the shared PlotHistory body works under both builds and the CLIENT_GUI-specific empty stub is gone.

Field mapping in HandlePacket mirrors CStatistics::GetPointsForUpdate so the same GetHistory + ComputeAverages code reads it back correctly: kBytes{Received,Sent} and kadNodesTotal are stored as session-rate * timestamp, so ComputeAverages's kValueRun / sTimestamp recovers the session-avg trend [0] verbatim. Per-point timestamps are reconstructed from the batch by stepping back from m_lastTimestamp at the 1 s scale we request in DoRequery.

No EC protocol changes

The ring fills from the existing EC_OP_GET_STATSGRAPHS stream — no new tags, no version bump, fully backward compatible with old daemons.

Test plan

• Statistics tab: graph stays continuous across Transfer / Search / Network switches; tree counters reflect current values on tab return.
• Network → Kad sub-tab: graph populates while sitting on Kad; no whole-graph blank when node count crosses a 50-step ceiling.
• Stats connection scope: no blank-out on nScale change.
• Monolithic amule.app builds cleanly with BUILD_MONOLITHIC=YES BUILD_REMOTEGUI=YES BUILD_DAEMON=YES; no regression in monolithic graph behavior (history machinery is the same code paths, just no longer #ifdef-gated).
• Long-session memory: listHR capped at 1800 records → bounded at ~150 KB total.