AssemblyAI alternative: migrate speech to text to Cartesia

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

## 1. Find the calls
Search for `api.assemblyai.com` or `ASSEMBLYAI_API_KEY`. The request to replace is `POST /v2/transcript`, then poll for `completed` or receive a webhook. List every call site before you edit anything.

## 2. Target
- Recorded files: `client.stt.transcribe(...)` in the `cartesia` SDK, or `POST https://api.cartesia.ai/stt`, with `model: "ink-whisper"`
- Live audio: `ink-2` over the STT WebSocket
- Headers: `X-API-Key` read from the `CARTESIA_API_KEY` environment variable, and `Cartesia-Version: 2026-08-14`
- Auth swap: AssemblyAI sends the key in `authorization` header holding the raw API key from `ASSEMBLYAI_API_KEY`. Replace it with `X-API-Key` from `CARTESIA_API_KEY` on the Cartesia path only, and keep `ASSEMBLYAI_API_KEY` for the AssemblyAI path. AssemblyAI takes the key with no `Bearer` prefix. Cartesia's SDK reads it from the client constructor and sends `X-API-Key` for you.
- Add `CARTESIA_API_KEY` wherever the project documents its environment (`.env.example`, README, deploy config). Never hardcode the key.
- Concurrency: AssemblyAI allows Pre-recorded: 5 parallel transcriptions on Free, 200+ on paid, with extra jobs queued. Streaming: 5 new sessions a minute on Free, 200 on paid. 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.
- Always pass `language`. Cartesia's batch call doesn't detect it and assumes `en`.

## 3. Field mapping, AssemblyAI → Cartesia
- `POST /v2/transcript`, then poll for `completed` or receive a webhook → `client.stt.transcribe(...)`, one call
- `authorization` header with the raw key → `Cartesia(api_key=...)`
- `audio_url`: a public URL or one from `/v2/upload` → `file`: an open binary file. Cartesia takes no URL input. Download audio that sits behind a URL and send the bytes.
- `speech_models` → `model`: `ink-whisper` for files, `ink-2` for live audio
- `language_code` → `language`. AssemblyAI detects the language when you leave it out. Cartesia batch assumes `en`, so set `language` on every request.

## 4. AssemblyAI fields with no Cartesia equivalent
- `keyterms_prompt`. What to do: Cartesia takes keyterms, up to 100, only on the Ink-2 WebSocket. `ink-whisper` batch has no keyterm parameter.
- `speaker_labels`. What to do: Cartesia returns no speaker labels on batch or streaming. If your product splits speakers, keep a separate diarization step.
- `prompt`. What to do: The batch call has no free-text prompt for steering formatting or style.
- `domain`: `medical-v1`. What to do: Cartesia has no medical mode. Transcribe your clinical recordings on both and compare word error rate before you move them.
- `webhook_url`. What to do: The transcript comes back in the response, so there is no job to be notified about. Call it from a worker if your request path can't wait on long files.

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

from cartesia import Cartesia

client = Cartesia(api_key=os.environ["CARTESIA_API_KEY"])

with open("call.wav", "rb") as audio_file:
    result = client.stt.transcribe(
        file=audio_file,
        model="ink-whisper",
        language="en",
        timestamp_granularities=["word"],
    )

print(result.text)
```

## 6. Roll out
1. Choose AssemblyAI or Cartesia per request from one config flag, defaulting to AssemblyAI.
2. Transcribe the same recordings with both and save the transcripts side by side. Compare the words, the word timestamps and the language on each.
3. Leave the AssemblyAI 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
- AssemblyAI submit transcript reference: https://www.assemblyai.com/docs/api-reference/transcripts/submit
- AssemblyAI models and languages: https://www.assemblyai.com/docs/getting-started/models
- AssemblyAI speaker diarization: https://www.assemblyai.com/docs/pre-recorded-audio/speaker-diarization
- Cartesia batch transcribe reference: https://docs.cartesia.ai/api-reference/stt/transcribe
- Cartesia keyterms guide: https://docs.cartesia.ai/use-the-api/stt/keyterms
- Cartesia speech-to-text docs: https://docs.cartesia.ai/use-the-api/stt/compare-endpoints

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