synth: frame-pace the mouse timeline so drags survive a slow/stalled frame

[borisbat]

Why

advance_mouse_timeline fired every event with t_ms <= now_ms (wall-clock), so on a coarse frame — the windows CI runner, where each test also eats ~10s of host spawn — a synth drag's press and destination-move collapse into one frame: imgui sees the button go down with the cursor already at the end, so the splitter/slider activates with a zero per-frame delta and never moves.

test_layout_helpers' real-drag subtest fails exactly this way on windows-latest (value/geometry unchanged) while passing on mac/ubuntu and on a fast local box.

Keys already solved this: imgui_key_chord / imgui_key_play are frame-paced (t_ms read as a frame index, one event-group per frame — see key_frame_paced). The mouse path never got it.

What

• imgui_live_core: mouse_frame_paced / mouse_frame_no; advance_mouse_timeline steps the threshold by frame index when paced (the existing lerp then interpolates one cursor delta per frame); MousePlayArgs.frame_paced opts in. Wall-clock callers (move_to / click_at / plain play / recordings) are byte-identical.
• imgui_playwright: drag() emits a frame-indexed timeline with frame_paced=true, so press → sweep → release each land on their own frame regardless of dt. drag_to() inherits.
• Also clears 4 pre-existing LINT012 surfaced by touching the file (_input on the status/stop [live_command] handlers, matching imgui_snapshot's pattern).

Validation

Reproduced locally by throttling the headless harness frame loop to 300ms/frame: test_layout_helpers drag FAIL → PASS after this change; drag_drop / io_synth_drag stay green at the same throttle; all pass un-throttled.

Proven on real windows CI via probe #183 (this fix on top of the windows-enable branch): --- PASS 'test_layout_helpers' (10.9s) on windows-latest.

Clicks and recordings (wall-clock) are untouched.

Note: the other windows-latest red, test_imgui_synth_ctrl_shortcuts, is a separate issue — the host goes unresponsive during its subtest 3 (snapshot + key_status both return null), the same family as the deferred test_docking_basic host crash. Deferred with that; not addressed here.

🤖 Generated with Claude Code