Stream batch audio uploads

[ComputelessComputer]

Summary:

• stream sync batch request bodies into temp files instead of extracting whole-body Bytes
• reuse the temp file path across provider retries and hyprnote routing attempts
• stream Deepgram raw uploads and OpenAI multipart file parts from disk

Verification:

• cargo check -p transcribe-proxy -p owhisper-client
• cargo test -p transcribe-proxy routes::batch
• cargo test -p owhisper-client adapter::deepgram::batch
• cargo test -p owhisper-client adapter::openai::batch
• git diff --check

---

Note

Medium Risk
Changes memory and I/O behavior on the main batch transcription path; incorrect streaming, temp file handling, or size limits could affect large uploads or retries, though logic is largely a refactor with tests added.

Overview
Batch sync transcription no longer buffers the full request body in memory. The transcribe-proxy batch handler accepts an Axum Body, streams chunks into a temp file (with MAX_BATCH_AUDIO_BODY_BYTES enforcement and clearer error responses), and passes that single file path through hyprnote routing, retries, and provider calls. Callback requests still buffer to Bytes for JSON parsing.

owhisper-client adds shared streaming_file_body / streaming_file_part helpers (reqwest stream feature). Deepgram batch uploads stream from disk with Content-Length; OpenAI multipart file parts stream from disk instead of loading the whole file into memory. Async callback sync fallback writes downloaded audio to a temp file before transcribing, matching the path-based API.

^{Reviewed by Cursor Bugbot for commit 418a5fd. Bugbot is set up for automated code reviews on this repo. Configure here.}

[netlify]

✅ Deploy Preview for old-char canceled.

Name	Link
🔨 Latest commit	418a5fd
🔍 Latest deploy log	https://app.netlify.com/projects/old-char/deploys/6a2bde56f520d40008ac31c5