Google Cloud TTS pricing and migration to Cartesia

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

## 1. Find the calls
Search for `texttospeech.googleapis.com` or `google.auth`. The request to replace is `POST texttospeech.googleapis.com/v1/text:synthesize`. 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: Google Cloud Text-to-Speech sends the key in `Authorization: Bearer` with an OAuth access token. Replace it with `X-API-Key` from `CARTESIA_API_KEY` on the Cartesia path only, and keep Google Cloud Text-to-Speech's key for the Google Cloud Text-to-Speech path. Google's authentication page covers Application Default Credentials, user credentials and service account impersonation. It lists no API key option for Text-to-Speech.
- Add `CARTESIA_API_KEY` wherever the project documents its environment (`.env.example`, README, deploy config). Never hardcode the key.
- Request size: Google Cloud Text-to-Speech caps a request at 5,000 bytes of text or SSML per request, so multi-byte scripts such as Japanese fit fewer characters. 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: Google Cloud Text-to-Speech allows Requests per minute per project: 1,000 on Neural2 and most voices, 500 on Studio, 200 on Chirp 3 HD, plus 100 concurrent streaming sessions. 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 Google Cloud Text-to-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-Neural2-F` (female), the voice in the code above → `db6b0ed5-d5d3-463d-ae85-518a07d3c2b4` (Skylar - Friendly Guide); `en-US-Chirp3-HD-Charon` (male) for reminders → `1df0b2fe-acb0-4484-8f32-9e231d9e7a90` (Harvey - Precise Liaison); `es-US-Neural2-A` (female) → `b4b8e2af-6139-466e-a93a-30c20d2e1fc5` (Fernanda - Friendly Guide); `hi-IN-Chirp3-HD-Kore` (female) → `e6b71342-48f1-4c70-a13a-d197b176ff24` (Sana - Friendly Concierge).

## 3. Field mapping, Google Cloud Text-to-Speech → Cartesia
- `POST texttospeech.googleapis.com/v1/text:synthesize` → `POST api.cartesia.ai/tts/bytes`
- Google Cloud credential (OAuth bearer token) → `X-API-Key` and `Cartesia-Version: 2026-08-14`. No cloud project or service account.
- `input.text` or `input.ssml` → `transcript`. Cartesia reads `<break>`, `<spell>`, `<emotion>`, `<speed>` and `<volume>` tags. Move other SSML into `generation_config` or a pronunciation dictionary.
- `voice.name`, such as `en-US-Neural2-F` → `voice`: `{"id": "<voice id>"}`. Google's pricing page files Neural2 and WaveNet under legacy models. Its current voices are Chirp 3 HD, which streams, and Gemini-TTS.
- `voice.languageCode` → `language`. Accepts a bare code such as `en` or a regional one such as `en-GB`.
- `audioConfig.audioEncoding` and `sampleRateHertz` → `output_format.encoding` and `sample_rate`. `MULAW` and `ALAW` map to `pcm_mulaw` and `pcm_alaw`. `OGG_OPUS` has no Cartesia container, so request `mp3` or raw PCM and transcode.
- `audioConfig.speakingRate` → `generation_config.speed` (0.6 to 1.5)
- `audioConfig.volumeGainDb` → `generation_config.volume` (0.5 to 2.0). Google takes decibels and Cartesia a 0.5 to 2.0 multiplier. Set it by ear.
- `enableTimePointing` with SSML `<mark>` (v1beta1) → TTS WebSocket with `add_timestamps`. Cartesia returns word timestamps and has no mark events.
- Response `audioContent` (base64) → Response body is the audio bytes

## 4. Google Cloud Text-to-Speech fields with no Cartesia equivalent
- `audioConfig.pitch`. What to do: Choose a voice whose natural pitch fits.

## 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 appointment is confirmed for Tuesday at 3 p.m."
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 Google Cloud Text-to-Speech or Cartesia per request from one config flag, defaulting to Google Cloud Text-to-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 Google Cloud Text-to-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
- Google Cloud text:synthesize reference: https://docs.cloud.google.com/text-to-speech/docs/reference/rest/v1/text/synthesize
- Google Cloud Text-to-Speech authentication: https://docs.cloud.google.com/text-to-speech/docs/authentication
- Google Cloud Text-to-Speech pricing: https://cloud.google.com/text-to-speech/pricing
- Google Cloud Text-to-Speech voice types: https://docs.cloud.google.com/text-to-speech/docs/list-voices-and-types
- 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.