Start a recording import. Returns an import_id and a single pre-signed
S3 url. Upload the file to url with PUT; that upload alone drives the
rest — the meeting is created and transcription starts automatically. There
is no separate "complete" call.

Flow:

1. POST /import (all fields optional: title, language, size, duration) → { import_id, url, expires_at }. If size is given and won't fit your remaining storage, the request fails fast with 403.
2. PUT {url} → upload the file bytes directly to S3. The Content-Type header is optional and need not match — the recording's type is detected from the file's contents, so a missing or generic Content-Type works fine.
3. Poll GET /import/{import_id}/status until status is ready, then use meeting_id

Concurrency: only one import start per user may be in flight at a time. Start
imports sequentially (await each before the next); there is no fixed cooldown —
once a call returns you may start the next immediately. A start that overlaps another
in-flight one returns 400 with error_code: ACTION_IN_PROCESS — retry that request.

Rate limit: 5 req/s (25 per 5s), enforced per account — shared across all your API keys. Exceeding returns 429.

Required scope: write