freecad: fckill + reloader watchdog for a stalled live session

[St0nefish]

Problem

The fclive reloader (a QTimer-based watcher) stalls, and there is no supported
way to force-recover a wedged live session. When it happens, the only recourse
today is the user hand-closing the GUI (Cmd-Q), which risks poisoning the next
launch because fcquit's cleanup never runs.

Migrated from freecad-plugin-gaps.md (bug/friction item 1), quoted in full:

fclive + fcrun clobber wedges the reloader. Running the headless
fcrun (or anything that writes the same WW_OUT/.FCStd) while a fclive
session holds that file open makes FreeCAD mark the doc dirty / detect an
external change, and the QTimer-based reloader silently stops firing (only
"rebuild #1" ever logged). Recovery is painful: fcquit can't complete
because the event loop is wedged, and it (correctly) refuses to kill.

• Fix ideas: SKILL.md should warn "never point fcrun at the same folder a
  live session owns; render/validate to a temp dir." Better: give fclive a
  way to self-validate (dump clashes/cutlist without a separate fcrun),
  and/or have the live doc suppress the "changed on disk" reload prompt. A
  stale .stop file also has to be removed by hand before relaunch.
• RECURRED without any clobbering — the QTimer reloader stalled on its own
  right after a big rebuild that dumps a long cutlist()/cutplan() into the
  Report View (Qt logs "table ... out of bounds, Resizing table model" as the
  last line each time). Strong smell that heavy per-rebuild Report View output
  stalls the event loop. Fixes: (a) don't run verbose cutlist/cutplan on every
  live rebuild (gate behind a flag / on-demand); (b) give the reloader a
  watchdog so a stalled QTimer restarts instead of silently dying; (c) make
  fcquit able to force-recover a wedged session without the user hand-closing
  it (Cmd-Q every time is rough).
• Need a fckill / force-recover tool. When the reloader wedges, the only
  recourse is the user hand-closing the GUI (Cmd-Q). There should be a
  supported force-kill that also cleans up afterward (clear FreeCAD's recovery
  breadcrumbs so the next launch doesn't hit the invisible "original file
  corrupted" modal, and remove the stale .stop/.snap files). Killing by
  hand today risks poisoning the next launch precisely because that cleanup
  isn't done.

Note: the SKILL.md warning about fcrun/fcrender clobbering a live session
(the first bullet above) has already shipped. What's still open is the tail
of this item — a watchdog for a QTimer reloader that stalls on its own (e.g.
after a heavy Report View dump), and a fckill force-recover tool that also
does the FreeCAD recovery-breadcrumb / stale .stop/.snap cleanup a manual
kill skips. This needs care around the event loop and the documented
"never kill, only clean shutdown" invariant fcquit currently upholds.

Source

Migrated from freecad-plugin-gaps.md in the freecad-woodworking project
(bug/friction item 1) — that file's "still open" recap calls this out
explicitly as: "reloader watchdog / fckill (#1 tail — event-loop and a
stated no-kill invariant, needs care)".