Integrate SpeechRecognitionService event handling

Add a private _subscribeToSpeechEvents() method called from startDictation() after the guard passes and permission is requested. Store the subscription as StreamSubscription? _speechEventSubscription and cancel it in dispose(). Use a switch expression on the event type for exhaustive event mapping — this prevents silent misses when new event types are added.

For SessionInterrupted, call await _partialRepo.save(fieldId, currentTranscript) inside the event handler before transitioning state; use a try/catch around the persist so a repository failure does not prevent the error state transition. PartialTranscriptionRepository injection at this stage can be a stub/placeholder until task-005 provides the full implementation — define the interface contract (abstract class) so tests can mock it. Avoid using scheduleMicrotask or Timer.run in the event handler; prefer direct state transitions to keep the state machine deterministic.

Unit tests using flutter_test with a fake/mock SpeechRecognitionService that exposes a StreamController. Test cases: (1) SpeechStarted → DictationRecording; (2) PartialResult in valid state → DictationRecording with updated interimTranscript; (3) PartialResult in invalid state (e.g., DictationIdle) is ignored; (4) FinalResult with non-empty text → DictationComplete; (5) FinalResult with empty text → DictationError(recognitionFailed); (6) RecognitionError → DictationError; (7) SessionInterrupted → PartialTranscriptionRepository.save() called then DictationError; (8) dispose() while recording → repository.save() called. Use expectLater with emitsInOrder for stream assertions. Minimum 90% line coverage on the event subscription handler code.