fix(cursor): system cursor swaps off the tick thread (#363)

[Maxaubert]

Closes #363.

The hitch recorder (#361) traced the zoom-in and zoom-out stutter to the system cursor swaps on the tick thread:

• Zoom-out restore: SPI_SETCURSORS reloads the whole cursor scheme from disk. 8 ms median, up to 90 ms under load. It froze the 1x landing frame.
• Blank at zoom-in and mid-zoom: 14 SetSystemCursor calls, about 10 ms (busy in cursor).

Change

• CursorBlanker runs its swaps in order on a Wind cursor swaps worker thread, at above normal priority.
• The zoom-out repaint nudge runs on the worker, after the restore.
• restoreSync() covers the input-panel clip nudge and shutdown. The crash filter is unchanged and still restores directly.
• The restore still reloads the live scheme, not cached copies, so animated cursors and cursor size changes survive.
• The txPace=2 composite pulse thread now runs at HIGHEST. Under build load it was starved (pulse-thread-late).
• Version 0.23.1.

Verified on the signed UIAccess build

• build.bat test: 569/569 pass. build.bat builds clean.

• Pointer check: 40 zoom cycles plus 5 in/out storms.

  • While zoomed, the system cursor was blank 40 of 40 times.
  • After every zoom-out the pointer was visible at 30 ms, 150 ms and 500 ms, and after every storm.

• Zoom-out teardown: 140 zoom-outs.

  	Average	Max
  Before	9.9 ms	35 ms in run 1, 90 ms in run 2
  After	3.4 ms	5 ms

  busy in cursor and zoom-out busy in activate hitches are gone.

Still open (follow-up)

• About 8 ms of loop-other between ticks right after a zoom-out. This is likely the message pump reacting to the cursor change, and needs a pump span.
• Zoom-in blocked in activate, 11 to 14 ms: DWM entering magnification.

🤖 Generated with Claude Code

https://claude.ai/code/session_01KPUNAWcwghXdHCApcKKjSG