feat: add Deepgram streaming transcription mode

[sjawhar]

Summary

Adds real-time streaming transcription via Deepgram's WebSocket API as a new whisper mode (streaming). Audio is transcribed as you speak, with results appearing after recording stops.

What's included

Core streaming client (src/transcribe/deepgram.rs):

• WebSocket client using the deepgram crate
• PCM audio encoding and streaming
• Final transcript extraction from Deepgram responses
• Drop impl that aborts the background WS task on cancel
• Error tracking for mid-stream disconnects
• 5-second timeout on stream finish to prevent daemon hangs
• UTF-8-safe transcript preview truncation

Daemon integration (src/daemon.rs):

• build_deepgram_config() with configurable sample rate passthrough
• Streaming recording paths in all 3 recording-start handlers (PTT, toggle, SIGUSR1)
• Model loading guard (skips local Whisper model prep in streaming mode)
• Audio capture cleanup on stream setup failure
• Rustls crypto provider init gated on streaming mode

Configuration (src/config.rs, src/cli.rs, src/main.rs):

• mode = "streaming" in config file
• --whisper-mode streaming CLI flag
• VOXTYPE_WHISPER_MODE environment variable
• --streaming-model, --streaming-endpoint, --streaming-api-key flags
• VOXTYPE_DEEPGRAM_API_KEY environment variable
• Streaming fields visible in voxtype config output (API key masked)

Documentation:

• docs/CONFIGURATION.md - streaming config reference with examples
• docs/USER_MANUAL.md - setup guide, usage, comparison with local mode
• docs/TROUBLESHOOTING.md - common streaming errors and fixes

Usage

# ~/.config/voxtype/config.toml
[whisper]
mode = "streaming"
streaming_api_key = "your-deepgram-api-key"

Or via environment:

export VOXTYPE_WHISPER_MODE=streaming
export VOXTYPE_DEEPGRAM_API_KEY="your-key"
voxtype daemon

Testing

• 538 tests pass (cargo test --lib)
• No changes to existing eager processing, local, remote, or CLI transcription modes
• New tests: UTF-8 truncation, sample rate propagation, config parsing