WellSaid alternative: move your WellSaid API to Cartesia

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

## 1. Find the calls
Search for `api.wellsaidlabs.com` or `WELLSAID_API_KEY`. The request to replace is `POST https://api.wellsaidlabs.com/v1/tts/stream`. 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: WellSaid sends the key in `X-Api-Key` header. Replace it with `X-API-Key` from `CARTESIA_API_KEY` on the Cartesia path only, and keep WellSaid's key for the WellSaid path. You request an API trial key after creating an account. Paid API plans go through sales.
- Add `CARTESIA_API_KEY` wherever the project documents its environment (`.env.example`, README, deploy config). Never hardcode the key.
- Request size: WellSaid caps a request at 1,000 characters per request. Cartesia publishes no per-request maximum, so keep any chunking the code already does and split at sentence boundaries; for long text, stream over the Cartesia TTS WebSocket.
- Concurrency: WellSaid allows 3 requests per second on the default API key. 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 WellSaid 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: Alana B. (3), clear US narration → `f0377496-2708-4cc9-b2f8-1b7fdb5e1a2a` (Elaine - Confident Guide); Wade C. (30), calm, deep US narration → `ed82c17b-4704-4d34-be43-5d19065acdf1` (Carl - Steady Storyteller); Ramona J. (5), friendly promo → `58fbaf73-d7de-4e82-a6b3-118180e7057c` (Janet - Sunny Speaker); Joe F. (28), promo with gravitas → `dbfa416f-d5c3-4006-854b-235ef6bdf4fd` (Damon - Commanding Narrator).

## 3. Field mapping, WellSaid → Cartesia
- `POST https://api.wellsaidlabs.com/v1/tts/stream` → `POST /tts/bytes`
- `X-Api-Key` header → `X-API-Key` and `Cartesia-Version: 2026-08-14`. Same header name. Add the version header.
- `speaker_id`: a numeric avatar ID such as `3` → `voice`: `{"id": "<voice id>"}`. Avatar IDs don't carry over. Pick a library voice per avatar.
- `text` (1,000 characters at most) → `transcript`
- `model`: `caruso` or `legacy` → `model_id`: `sonic-3.6`. WellSaid falls back to `legacy` when `model` is missing. Always send `model_id` on Cartesia.
- `audio_configs.file_format`: `mp3`, `wav` or `s16le` → `output_format.container`: `mp3`, `wav` or `raw` with `pcm_s16le`
- `audio_configs.sample_rate` (8000 to 96000) → `output_format.sample_rate` (8000 to 48000). Rates above 48000 have no equivalent. Resample if your pipeline needs them.
- `X-Enable-SSML: true` (limited SSML) → SSML-style tags in `transcript`: `break`, `spell`, `emotion`, `speed`, `volume`. No header needed. Check each tag you send today against [Cartesia's list](https://docs.cartesia.ai/build-with-cartesia/capability-guides/ssml-tags).
- `library_ids` (replacement libraries) → `pronunciation_dict_id` (one dictionary). Copy each library's replacements into one Cartesia dictionary.

## 4. 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 = "Module four covers the new expense policy. Receipts over fifty dollars need a manager's approval."
VOICE_ID = "f0377496-2708-4cc9-b2f8-1b7fdb5e1a2a"  # 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": "mp3",
            "sample_rate": 44100,
            "bit_rate": 128000,
        },
    },
)
response.raise_for_status()
open("out.mp3", "wb").write(response.content)
```

## 5. Roll out
1. Choose WellSaid or Cartesia per request from one config flag, defaulting to WellSaid.
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 WellSaid 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
- WellSaid TTS stream API reference: https://docs.wellsaidlabs.com/reference/ttsstream
- WellSaid getting started: https://docs.wellsaidlabs.com/docs/getting-started-1
- WellSaid model selection: https://docs.wellsaidlabs.com/reference/model-selection-with-the-api
- WellSaid voice avatars: https://docs.wellsaidlabs.com/reference/available-voice-avatars
- 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.