Continue MutationObserver batches after stale records

[OskarEichler]

Summary:

MutationObserverManager intentionally tolerates stale native records after an observer disconnects, but currently returns from the complete notification pass and drops callbacks for every later active observer in that batch. Skip only the stale observer, matching the equivalent ResizeObserver path, and add a deterministic queued-record regression.

Fixes #58236.

Changelog:

[GENERAL] [FIXED] - Continue dispatching active MutationObserver callbacks after a stale observer record.

Test Plan:

• Exact upstream focused Fantom run: 27 existing tests passed and the new stale-batch regression failed.
• Fixed focused Fantom run: 28/28 passed across two suites.
• The regression unregisters the stale observer at the manager boundary to preserve its already queued native record and model the documented commit-to-mount race.
• Fresh Flow check: 0 errors.
• Targeted no-ignore ESLint, Prettier, and git diff --check passed.

No UI change; screenshots are not applicable.