Avoid event-loop reentry in window height checks

[adkins17]

Summary

• remove the nested Tk update from ::potato::countWindowHeight
• prevent window-height/NAWS measurement from re-entering queued UI, input, timer, or socket callbacks
• add a stock Tcl/Tk regression test and PowerShell runner
• include a 500-call latency measurement in the test output

Reproducer

The test queues an unrelated callback immediately before calling countWindowHeight.
On unmodified stock Potato, the callback runs inside the function because update
processes the complete Tk event queue. With this patch, it remains pending until the
normal event loop resumes.

Results

Tcl/Tk 8.6.18, Windows, upstream baseline 46d18387bd6b156b85176fa588a60991cbba6235:

Build	Re-entry assertion	p50	p95	max
Stock	FAIL	0.011 ms	0.060 ms	0.600 ms
Patched fresh worktree	PASS	0.005 ms	0.006 ms	0.237 ms

The primary correction is eliminating nested event-loop processing; the microbenchmark
also showed a roughly 90 percent p95 reduction.

Test

./tests/Test-WindowHeightLatency.ps1 -Runtime C:/path/to/tclkit-8.6.18-win64-intel-tk.exe

Fresh-worktree validation passed:

• patch apply check
• patch reverse check
• PowerShell parser
• all window-height regression assertions

This change is based directly on stock Potato and has no dependency on split-screen,
packet-debugger, telemetry, or other custom-client code.