Fix timer issues

[danyabdo94]

Fix timer race conditions, double elapsed deduction, and pauseFor behaviour

Summary

This PR fixes 6 bugs across _setupListenAndPause, _notifyResults, _stop, _shutdownListener, and initialize. The issues range from a silent misconfiguration to a critical double-subtraction that caused the listen/pause timer to fire far too early.

---

Changes

_setupListenAndPause — double subtraction of _elapsedListenMillis (critical)
In the else branch (both pauseFor and listenFor non-null), listenFor had already had _elapsedListenMillis subtracted in the adjustment block above. The minMillis calculation then subtracted it a second time, producing a value far smaller than intended — potentially negative. The redundant subtraction has been removed and the result is now clamped to 0 to prevent a negative Duration being passed to Timer.

_setupListenAndPause — orphaned timer on re-entry
If _setupListenAndPause was called while _listenTimer was still active (e.g. via _stopOnPauseOrListen → _setupListenAndPause, or changePauseFor), a second timer was created without cancelling the first. Both would eventually fire _stopOnPauseOrListen. The existing timer is now cancelled at the top of the method before any new one is created.

_notifyResults — pauseFor timer reset when _partialResults is false (critical)
The _lastSpeechEventAt timestamp (which drives the pauseFor countdown) was being updated before the early-return check for _partialResults. This meant every interim result silently reset the pause timer even when the caller had opted out of partial results, making pauseFor effectively non-functional in that mode. The early-return has been moved above the timestamp update.

_stop — _notifyFinalTimer not cancelled before reassignment
If _stop was called while a _notifyFinalTimer was already pending, a second timer was created alongside the first. Both would fire _onFinalTimeout. A defensive _notifyFinalTimer?.cancel() has been added immediately before the reassignment.

initialize — empty if block silently ignored invalid finalTimeout
The block if (finalTimeout <= _minFinalTimeout) {} was a no-op, so passing a sub-threshold finalTimeout would set _finalTimeout to the invalid value and silently disable the final timeout logic. The block now clamps _finalTimeout to _minFinalTimeout.

_shutdownListener — duplicate _listenTimer = null
A copy-paste leftover left _listenTimer = null written twice in sequence. The duplicate line has been removed.