Fix textarea undo crash

[guillaumelgf]

This PR fixes #6686

Summary

Fixes a TextArea undo crash that can happen after undoing a large multiline paste.

During undo, the document may already be restored to its shorter previous state while the cursor location still temporarily points into the pasted text. If scrollbar refresh tries to scroll that stale cursor location into view, WrappedDocument.location_to_offset() can
raise because the row no longer exists.

This change clamps the cursor location used for the visual offset calculation, without mutating the actual selection. The undo flow can then finish restoring the selection normally.

[goransh-walia]

This looks good to me at a glance — thanks for putting it together.

[Phantom8420]

The approach is reasonable — _recompute_cursor_offset only maintains a visual cache (self._cursor_offset), so clamping the location used for that calculation without touching self.selection is proportionate. clamp_visitable is already the established idiom for this (_clamp_selection, move_cursor_relative both use it), so this isn't introducing a new pattern.

A few things before this merges:

1. CHANGELOG.md isn't updated — the PR checklist requires it. A ### Fixed entry with the PR link, matching the existing format, is missing.
2. Would you rather fix the root ordering instead — the selection getting restored to a stale location while the document is already shortened during undo — so _recompute_cursor_offset is never called with an invalid location in the first place? The clamp is a fine safety net either way, just flagging the alternative.
3. The test asserts selection == Selection.cursor((0, 0)) after undo but doesn't check that the resulting cursor offset/visual position ends up correct once the undo flow settles. If a later _recompute_cursor_offset call with the now-restored location fixes it up, an assertion for that (or a comment explaining why it's guaranteed) would lock it in rather than just proving the crash is gone.

Nit: a comment on why the test needs size=(120, 40) + height: 8 would help — presumably to force wrapping/scrolling so the scrollbar-refresh path that triggers the crash actually runs.