Migrate Amazon Polly SynthesizeSpeech to Cartesia

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

## 1. Find the calls
Search for `boto3`. The request to replace is `SynthesizeSpeech` through boto3, signed with AWS credentials. 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: Amazon Polly sends the key in AWS Signature Version 4, which boto3 signs for you from `AWS_ACCESS_KEY_ID`. Replace it with `X-API-Key` from `CARTESIA_API_KEY` on the Cartesia path only, and keep `AWS_ACCESS_KEY_ID` for the Amazon Polly path. boto3 also reads `AWS_SECRET_ACCESS_KEY` or an attached role, and the IAM policy has to allow each `polly:` action you call. Cartesia needs only the key.
- Add `CARTESIA_API_KEY` wherever the project documents its environment (`.env.example`, README, deploy config). Never hardcode the key.
- Request size: Amazon Polly caps a request at 3,000 billed characters (6,000 in total) per SynthesizeSpeech call; 100,000 billed through StartSpeechSynthesisTask. 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: Amazon Polly allows Neural: 8 requests a second, burst 10, up to 18 at once. Standard: 80 a second and 80 at once. Generative and long-form: 26 at once. 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 Amazon Polly 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: Joanna (US English, female) → `db6b0ed5-d5d3-463d-ae85-518a07d3c2b4` (Skylar - Friendly Guide); Matthew (US English, male) → `47c38ca4-5f35-497b-b1a3-415245fb35e1` (Daniel - Modern Assistant); Amy (British English, female) → `62ae83ad-4f6a-430b-af41-a9bede9286ca` (Gemma - Decisive Agent); Brian (British English, male) → `4bc3cb8c-adb9-4bb8-b5d5-cbbef950b991` (George - Composed Consultant).

## 3. Field mapping, Amazon Polly → Cartesia
- `SynthesizeSpeech` through boto3, signed with AWS credentials → `POST /tts/bytes` with `X-API-Key` and `Cartesia-Version: 2026-08-14`. One API key replaces the IAM policy and request signing.
- `Text`, with `TextType: ssml` for markup → `transcript`. `<break time="1s"/>` works as written, and Cartesia needs no `<speak>` wrapper. `<prosody>`, `<say-as>`, `<phoneme>` and `<mark>` have no effect; [SSML tags](https://docs.cartesia.ai/build-with-cartesia/capability-guides/ssml-tags) lists the five Cartesia reads.
- `VoiceId`, such as `Joanna` → `voice`: `{"id": "<voice id>"}`
- `Engine`: `standard`, `neural`, `long-form` or `generative` → `model_id`: `sonic-3.6`. Every Cartesia voice runs on the same model, so there is no engine to pick per voice.
- `LanguageCode`, such as `en-US` → `language`: `en` or `en-US`. Polly reads it only for bilingual voices. Cartesia wants it on every request.
- `OutputFormat`: `mp3`, `pcm`, `mulaw` or `alaw` → `output_format.container` and `encoding`. `mp3` is the `mp3` container. `pcm` is `raw` with `pcm_s16le`, `mulaw` is `pcm_mulaw` and `alaw` is `pcm_alaw`.
- `SampleRate` → `output_format.sample_rate`. Any rate from 8000 to 48000 Hz.
- `OutputFormat: json` with `SpeechMarkTypes` (`word`, `sentence`, `viseme`, `ssml`) → `add_timestamps` on the TTS WebSocket. The WebSocket returns word timestamps, and phoneme timestamps with `add_phoneme_timestamps`. `/tts/bytes` returns neither, and Cartesia has no sentence, viseme or SSML marks.
- `LexiconNames`, up to 5 per request → `pronunciation_dict_id`, one per request. Merge the lexicons you stack today into a single dictionary.
- `StartSpeechSynthesisStream` (generative engine) → The TTS WebSocket. Send text as it arrives and read audio chunks back; the [WebSocket reference](https://docs.cartesia.ai/api-reference/tts/websocket) has the message format.

## 4. Amazon Polly fields with no Cartesia equivalent
- `OutputFormat`: `ogg_vorbis` or `ogg_opus`. What to do: Ask for `mp3` or raw PCM and encode Ogg yourself.

## 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 package will arrive between two and four this afternoon."
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": "raw",
            "encoding": "pcm_s16le",
            "sample_rate": 16000,
        },
    },
)
response.raise_for_status()
open("out.pcm", "wb").write(response.content)
```

## 6. Roll out
1. Choose Amazon Polly or Cartesia per request from one config flag, defaulting to Amazon Polly.
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 Amazon Polly 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
- Amazon Polly SynthesizeSpeech reference: https://docs.aws.amazon.com/polly/latest/dg/API_SynthesizeSpeech.html
- Amazon Polly StartSpeechSynthesisStream reference: https://docs.aws.amazon.com/polly/latest/dg/API_StartSpeechSynthesisStream.html
- 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
- Amazon Polly quotas: https://docs.aws.amazon.com/polly/latest/dg/limits.html

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