feat(frontend): improve audio recorder lifecycle and STT integration

[chriswritescode-dev]

Enhances the audio recording infrastructure for speech-to-text with improved lifecycle management and comprehensive test coverage.

Changes

Audio Recorder Improvements

• Added prepare() method to initialize AudioContext in suspended state without acquiring microphone
• Implemented lifecycle management: prepare → start → stop → releaseStream → dispose
• AudioContext now suspends on stop instead of closing, enabling efficient restarts
• MediaStream tracks are properly disabled on stop and released on demand
• Idempotent prepare() prevents duplicate worklet loading

STT Hook Updates

• Improved integration with new audio recorder lifecycle
• Better error handling and state management
• Added startup timeout handling for recorder initialization

Testing

• Added comprehensive lifecycle tests for AudioRecorder
• Mocked AudioContext, AudioWorkletNode, and MediaStream APIs
• Tests cover: prepare, start, stop, releaseStream, dispose, and recovery scenarios

Build Fixes

• Updated zod to ^4.4.3 for compatibility with @hookform/resolvers
• Added vite optimizeDeps configuration for zod and @hookform/resolvers

Testing

• Verified audio recording flow works correctly
• All new tests pass
• Frontend dev server starts without errors