v3.5b1: Volume timing fixes, TTS cache improvements, diagnostics

Bug Fixes

Volume Restoration Timing

• Fixed volume being restored too early when using chime or normalize features
• Volume restoration now correctly waits for full playback duration
• Added platform-specific timing buffers (Sonos: 500ms, Cast: 700ms)

Chime Configuration

• Fixed chime settings not being applied from entity configuration
• Service now reads chime/normalize settings from TTS entity subentry config

Voice Change Bug

• Added default_options property to TTS entity
• Voice/model/speed changes in config now properly invalidate HA's TTS cache

New Features

Duration Metadata with Mutagen

• Added reliable MP3 metadata embedding using mutagen library
• Duration stored in ID3 TXXX:duration_ms tag

In-Memory Duration Cache

• Added MESSAGE_DURATIONS_KEY for fast duration lookups
• Shared cache across all TTS entities

Diagnostics Support

• New diagnostics.py for troubleshooting
• Exports config entry data (API key redacted)

Code Quality

• Fixed deprecated asyncio.get_event_loop() calls
• Fixed bare except: clauses
• Removed duplicate constant definitions
• Added type annotations