Move your Fish Audio TTS API to Cartesia

Move this codebase from Fish Audio to Cartesia for text to speech. Keep Fish Audio working behind a flag until I switch over.

December 31, 2026: S1 retires. Fish Audio lists `s1` as deprecated. Requests that send `model: s1` need a new model before the end of the year either way. (https://docs.fish.audio/developer-guide/models-pricing/models-overview)

## 1. Find the calls
Search for `api.fish.audio` or `FISH_API_KEY`. The request to replace is `POST https://api.fish.audio/v1/tts`. List every call site before you edit anything.

## 2. Target
- Request: `POST https://api.cartesia.ai/tts/bytes`
- Model: `model_id: "sonic-3.6"`
- Headers: `X-API-Key` read from the `CARTESIA_API_KEY` environment variable, and `Cartesia-Version: 2026-08-14`
- Auth swap: Fish Audio sends the key in `Authorization: Bearer` token from `FISH_API_KEY`. Replace it with `X-API-Key` from `CARTESIA_API_KEY` on the Cartesia path only, and keep `FISH_API_KEY` for the Fish Audio path. Fish Audio shows a new key once, on its developer page. Copy it before you close the dialog.
- Add `CARTESIA_API_KEY` wherever the project documents its environment (`.env.example`, README, deploy config). Never hardcode the key.
- Concurrency: Fish Audio allows 5 concurrent requests under $100 paid, 15 from $100, 50 from $1,000. Look up the Cartesia plan's concurrent request limit at https://www.cartesia.ai/pricing, cap parallel requests at it, and retry rate-limit errors with backoff.
- Voice IDs don't carry over. Keep each voice ID in one named constant or config value, put a placeholder where each Fish Audio voice was, and list them for me so I can pick replacements at https://www.cartesia.ai/voices.
- Suggested starting points to put in those placeholders, for me to confirm by ear: Expressive narrator for video content → `a33f7a4c-100f-41cf-a1fd-5822e8fc253f` (Lauren - Lively Narrator); Upbeat promo or ad read → `710feaa3-b550-42f3-b3eb-6f37f2a7cc0a` (Tanner - Upbeat Assistant); Friendly in-game companion → `6ccbfb76-1fc6-48f7-b71d-91ac6298247b` (Tessa - Kind Companion); Slow, calm long-form read → `98a34ef2-2140-4c28-9c71-663dc4dd7022` (Clyde - Calm Narrator).

## 3. Field mapping, Fish Audio → Cartesia
- `POST https://api.fish.audio/v1/tts` → `POST /tts/bytes`
- `Authorization: Bearer $FISH_API_KEY` → `X-API-Key` and `Cartesia-Version: 2026-08-14`
- `model` header: `s2.1-pro`, `s2-pro`, `s1` or `drama-3` → `model_id`: `sonic-3.6` in the body
- `text` → `transcript`
- `reference_id` → `voice`: `{"id": "<voice id>"}`. Fish Audio voice models don't export. Pick a library voice or clone the source recording.
- `format`: `wav`, `pcm` or `mp3` → `output_format`: `container` `wav`, `raw` or `mp3`. Fish Audio's `pcm` is 16-bit mono. Set `raw` with `pcm_s16le` to match it.
- `sample_rate` → `output_format.sample_rate`
- `mp3_bitrate`: 64, 128 or 192 → `output_format.bit_rate`: 32000 to 192000. Cartesia counts bits per second, so 128 becomes 128000.
- `prosody.speed` (0.5 to 2.0) → `generation_config.speed` (0.6 to 1.5). Cap anything under 0.6 or over 1.5 at the nearest end.
- `prosody.volume` (decibels) → `generation_config.volume` (0.5 to 2.0). Cartesia uses a multiplier, not decibels. Match loudness by listening.
- Emotion tags in the text, `[happy]` on S2 or `(happy)` on S1 → `generation_config.emotion`, or an `emotion` tag in the transcript. Cartesia accepts a fixed list of emotions. Free-form cues such as `[whispers sweetly]` need rewording.
- `pronunciation_dictionary` (up to 3) → `pronunciation_dict_id` (one dictionary). Merge your dictionaries into one before uploading.

## 4. Fish Audio fields with no Cartesia equivalent
- `references`: inline audio and transcript, MessagePack only. What to do: Clone the clip once with `client.voices.clone(clip=..., language=..., name=...)` and reuse the ID.
- `format`: `opus`. What to do: Request `mp3` or raw PCM and transcode to Opus.
- `latency`, `chunk_length`, `temperature`, `top_p`. What to do: Drop them.
- `<|speaker:0|>` multi-speaker tags. What to do: Send each speaker's lines as their own request and join the audio.

## 5. Reference request
Match the language and HTTP client the codebase already uses. This is the same request in Python:
```python
import os
import requests

text = "Your character reached level ten. A new quest is waiting at the harbor."
VOICE_ID = "6ccbfb76-1fc6-48f7-b71d-91ac6298247b"  # from /voices

response = requests.post(
    "https://api.cartesia.ai/tts/bytes",
    headers={
        "X-API-Key": os.environ["CARTESIA_API_KEY"],
        "Cartesia-Version": "2026-08-14",
    },
    json={
        "model_id": "sonic-3.6",
        "transcript": text,
        "voice": {"id": VOICE_ID},
        "language": "en",
        "output_format": {
            "container": "wav",
            "encoding": "pcm_s16le",
            "sample_rate": 44100,
        },
    },
)
response.raise_for_status()
open("out.wav", "wb").write(response.content)
```

## 6. Roll out
1. Choose Fish Audio or Cartesia per request from one config flag, defaulting to Fish Audio.
2. Run the same scripts through both and save the clips side by side for me to listen to. Check that the container, encoding and sample rate match what downstream code expects.
3. Leave the Fish Audio client, call and key in place so rollback is flipping the flag. Don't delete them in this change.
4. Run the existing tests, and add one that checks the Cartesia request body.

## Docs
- Fish Audio text-to-speech API reference: https://docs.fish.audio/api-reference/endpoint/openapi-v1/text-to-speech
- Fish Audio quickstart: https://docs.fish.audio/developer-guide/getting-started/quickstart
- Fish Audio voice cloning best practices: https://docs.fish.audio/developer-guide/best-practices/voice-cloning
- Fish Audio emotion tags: https://docs.fish.audio/developer-guide/core-features/emotions
- Cartesia TTS bytes API reference: https://docs.cartesia.ai/api-reference/tts/bytes
- Cartesia SSML tags: https://docs.cartesia.ai/build-with-cartesia/capability-guides/ssml-tags

When you're done, tell me which files changed, which fields you couldn't map, and how to flip the flag.