Migrate Azure TTS from SSML to Cartesia

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

## 1. Find the calls
Search for `SPEECH_KEY`. The request to replace is `POST https://{region}.tts.speech.microsoft.com/cognitiveservices/v1`. 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: Azure Speech sends the key in `Ocp-Apim-Subscription-Key` with the resource key, or `Authorization: Bearer` with a token from `issueToken` from `SPEECH_KEY`. Replace it with `X-API-Key` from `CARTESIA_API_KEY` on the Cartesia path only, and keep `SPEECH_KEY` for the Azure Speech path. A token from `issueToken` is valid for 10 minutes. Microsoft's quickstart also sets `SPEECH_REGION` next to the key.
- Add `CARTESIA_API_KEY` wherever the project documents its environment (`.env.example`, README, deploy config). Never hardcode the key.
- Request size: Azure Speech caps a request at 10 minutes of audio per request, and Azure cuts longer output at 10 minutes; SSML holds up to 50 `<voice>` and `<audio>` tags. 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: Azure Speech allows 20 transactions per 60 seconds on F0, not adjustable; 30 per second by default on S0, adjustable up to 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 Azure Speech 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: `en-US-ChristopherNeural` (male), the voice in the code above → `68fb6747-b6ea-4c44-a18e-4e29921424d3` (Derek - Deep Advisor); `en-US-AvaMultilingualNeural` (female) → `55deba52-bc73-4481-ab69-9c8831c8a7c3` (Camille - Friendly Expert); `de-DE-KatjaNeural` (female) → `9b4d08b6-0494-4301-ab92-9150f4ee2718` (Marlene - Elegant Speaker).

## 3. Field mapping, Azure Speech → Cartesia
- `POST https://{region}.tts.speech.microsoft.com/cognitiveservices/v1` → `POST https://api.cartesia.ai/tts/bytes`
- `Ocp-Apim-Subscription-Key` or `Authorization: Bearer` → `X-API-Key` and `Cartesia-Version: 2026-08-14`
- `Content-Type: application/ssml+xml` → `Content-Type: application/json`
- SSML `<voice name="en-US-ChristopherNeural">` → `voice`: `{"id": "<voice id>"}`
- Text inside the SSML body → `transcript`. SSML-style `<break>`, `<spell>`, `<emotion>`, `<speed>` and `<volume>` tags work inside it.
- SSML `xml:lang` → `language`
- `<prosody rate>` and `<prosody volume>` → `generation_config.speed` (0.6 to 1.5) and `volume` (0.5 to 2.0)
- `<mstts:express-as style="cheerful">` → `generation_config.emotion` or an `<emotion>` tag. `styledegree` and `role` have no equivalent. Map each style to the nearest emotion and listen.
- `<phoneme>` and `<lexicon>` → `pronunciation_dict_id`
- `X-Microsoft-OutputFormat: raw-8khz-8bit-mono-mulaw` → `output_format`: `{"container": "raw", "encoding": "pcm_mulaw", "sample_rate": 8000}`
- `X-Microsoft-OutputFormat: riff-24khz-16bit-mono-pcm` → `output_format`: `{"container": "wav", "encoding": "pcm_s16le", "sample_rate": 24000}`
- Speech SDK `WordBoundary` events → TTS WebSocket with `add_timestamps`. `/tts/bytes` returns no timing. Sentence boundaries have no equivalent.
- 10 minutes of audio per REST request → Split long scripts across several requests, or stream them over the WebSocket

## 4. Azure Speech fields with no Cartesia equivalent
- `<prosody pitch>` and `<prosody contour>`. What to do: Pick a voice that already sits at the pitch you want.
- Speech SDK `VisemeReceived` events and `<mstts:viseme>`. What to do: If you drive lip sync yourself, phoneme timestamps from `add_phoneme_timestamps` are the closest input.

## 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 = "We've received your claim. A specialist will call you within two business days."
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": 24000,
        },
    },
)
response.raise_for_status()
open("out.wav", "wb").write(response.content)
```

## 6. Roll out
1. Choose Azure Speech or Cartesia per request from one config flag, defaulting to Azure Speech.
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 Azure Speech 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
- Azure text to speech REST API reference: https://learn.microsoft.com/en-us/azure/ai-services/speech-service/rest-text-to-speech
- Azure SSML voice and style reference: https://learn.microsoft.com/en-us/azure/ai-services/speech-service/speech-synthesis-markup-voice
- Azure Speech quotas and limits: https://learn.microsoft.com/en-us/azure/ai-services/speech-service/speech-services-quotas-and-limits
- Azure language and voice support: https://learn.microsoft.com/en-us/azure/ai-services/speech-service/language-support?tabs=tts
- Azure text to speech quickstart (Python): https://learn.microsoft.com/en-us/azure/ai-services/speech-service/get-started-text-to-speech?pivots=programming-language-python
- Azure Speech release notes: https://learn.microsoft.com/en-us/azure/ai-services/speech-service/releasenotes
- 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.