tts --stream --with-timestamps: route to /stream/with-timestamps instead of hard-erroring

[longevityboris]

Context

src/commands/tts.rs rejects --stream + --with-timestamps as incompatible, but the OpenAPI spec has a dedicated endpoint for exactly this combination: POST /v1/text-to-speech/{voice_id}/stream/with-timestamps (operationId text_to_speech_stream_with_timestamps).

The dialogue command already handles the streaming + timestamps combination via NDJSON; TTS should too.

Use case

Low-latency karaoke / live subtitle generation. Users currently have to choose between latency (stream) OR per-char alignment (with-timestamps) but not both.

What to do

• When both flags are set, route to /v1/text-to-speech/{voice_id}/stream/with-timestamps.
• Parse the NDJSON response (one chunk per audio frame, each with alignment), writing audio bytes to --output and alignment to --save-timestamps.
• Update TTS_HELP + agent-info to reflect that the combo now works.

Files

• src/commands/tts.rs
• src/help.rs (TTS_HELP)
• src/commands/agent_info.rs

Acceptance

• tts "hello" --stream --with-timestamps -o out.mp3 --save-timestamps t.jsonl writes both files.
• Integration test mocks the NDJSON endpoint.