Replaces every mousedown/mousemove/mouseup/mouseenter/mouseleave
listener in the slides editor module with the pointer-event
equivalent. PointerEvent is a strict superset of MouseEvent (it
inherits from it in TS), and browsers synthesize pointer events
from both mouse and touch — so this rename costs nothing on
desktop but is the prerequisite for touch drag on iOS Safari,
which never synthesizes mousemove during a touch drag (only
mousedown/mouseup for taps). Pen tablets and stylus input get
supported as a side-effect.

Scope is the slides editor module only (editor.ts, thumbnail-
panel.ts, context-menu.ts, layout-picker.ts). The text-box-editor
delegates click handling to the docs TextEditor, which still uses
mouse events; tests that target it keep MouseEvent dispatches.
The presenter module (read-only playback) is also out of scope.

Phase B spike findings that drove this:
docs/tasks/active/20260517-slides-mobile-edit-lessons.md

The mobile shell needs fingertips to land on 8px handles reliably.
Add a `tolerance` parameter to `handleHitTest` (default 0 — desktop
unchanged) that expands each handle's hit rectangle by that many
pixels on every side. `SlidesEditor` accepts the matching
`touchHandleTolerance` option and forwards it to every hit-test
call.

22px expansion around 9 handles on a small selection makes their
hit rectangles overlap, so a flat first-match-wins iteration picked
by DOM order and felt non-deterministic from a fingertip. Switch
to closest-center: among all handles whose expanded rect contains
the point, pick the one whose center is nearest. Two new unit tests
cover the overlap path and confirm desktop tolerance=0 behavior
when only one rect matches stays equivalent.

Adds a `mode: 'view' | 'edit'` prop to MobileSlidesView, defaulting
to 'edit'. The 'view' branch keeps Phase A's read-only SlideRenderer
path with swipe nav. The 'edit' branch builds a canvas + overlay
and mounts the desktop SlidesEditor against them with
touchHandleTolerance=22 forwarded — touch drag/resize/rotate work
out of the box via the Pointer Events migration in the previous
commit (one-back).

Two iOS-specific guards on the canvas-host: touch-action:none
gives the editor sole ownership of horizontal pointer (no
competition from browser swipe nav, double-tap zoom, pinch);
-webkit-touch-callout + user-select kill the long-press callout
preview that contextmenu blocking does NOT cover. slides-detail
passes mode="edit" explicitly. Permission-gated 'view' fallback
for shared-link viewers is the Task 4 wiring.

Phase B Task 1b — see
docs/tasks/active/20260517-slides-mobile-edit-todo.md.

Three findings from the sim smoke pass: (1) flat 22px hit tolerance
around 9 close-packed handles needed closest-center disambiguation
to feel deterministic — fix in the previous slides commit. (2) iOS
Hangul IME composition in the docs text-box mounted from slides
behaves differently from desktop; deferred to Task 2 (text-edit
PR) since that's where docs-side IME work belongs. (3) The
_spike-login.html cookie installer is the repeatable trick for
dropping the iPhone sim into an authenticated session — local-only,
never committed, recreate per spike.

Self-review caught two issues in the new edit-mode mount:

1. **Listener leak (Critical):** the cleanup dropped editorRef but
   never called editor.detach(), so every mount cycle leaked the
   editor's document-level pointer/key listeners, any in-flight RAF,
   and an active text-box. Desktop slides-view.tsx:479 already
   tears down the same way — mobile now matches.

2. **Stale-closure setState (Important):** the onCurrentSlideChange
   callback compared the editor's reported slide id against
   currentSlideId captured in closure at mount time, which never
   refreshed because the effect intentionally omits currentSlideId
   from its dep list (so footer-arrow taps don't tear down the
   editor). Switch to the prev-state reducer form so we always
   compare against the latest React state.

Code review surfaced two non-blocking items kept out of this PR
but worth recording for the next person:

- The `[data-handle] { pointer-events: auto !important }` style is
  injected as a global selector by BOTH desktop (slides-view.tsx)
  and mobile (mobile-slides-view.tsx). The mobile mount inherits
  the desktop pattern verbatim; isMobile branches so they never
  fight. A follow-up should scope both. Logged in lessons.
- jsdom defaults PointerEvent's pointerId to 0; real browsers use
  1+. Not a present bug because the editor doesn't read pointerId,
  but a trap for any future PR that adds setPointerCapture. Logged.

Also drops a confusing tie-break paragraph in the closest-center
overlap test — the test uses biased points that make the tie
analysis irrelevant.