refactor(api-proxy): read the image from a buffered call

[rickstaa]

Blocked on livepeer/livepeer-python-gateway#51 and #52 landing in ja/live-runner. The example pins that branch, so no pyproject.toml change is needed; mark ready and merge once both are in.

One bounded call returning a 120 KB image has nothing to stream. stream=True was there because call_runner's buffered path assumed a JSON object and raised on image/jpeg. With #51 the body arrives in result.content, so the async with / aiter_bytes() block collapses to one line.

One thing worth knowing: the buffered path bounds the whole request with call_runner's 5s default, which streaming did not (open_stream sets sock_read=None). Dropping stream=True without a timeout fails after ~5s with failed to reach endpoint and no mention of a timeout, so the call passes timeout=180.

The nginx Accept pin stays. It is not a workaround for the SDK bug: the runner advertises an image at a fixed price, so the response format is its choice, and the pin still holds for callers that are not this SDK or are on an older pin.

Tested

End to end on the paid path against the stacked SDK branch (#51 + #52), with the pin in place: a real Hugging Face call returned a 1024x1024 image/jpeg (56313 bytes) into result.content. The temporary SDK pin used for that run was reverted before committing.

Closes #47

---

Unblocked and re-verified 2026-08-12. Both SDK PRs landed on 2026-07-31 and are now in livepeer-python-gateway's main, which this repo pins after #66, so no pyproject.toml change is needed:

• feat: return raw bytes from call_runner for non-JSON responses livepeer-python-gateway#51 — return raw bytes from call_runner for non-JSON responses
• fix(call_runner): stop asking apps for JSON livepeer-python-gateway#52 — stop asking apps for JSON

Rebased onto main and re-ran end to end against SDK main (a0b9129) rather than the stacked branch used for the original test. A real Hugging Face call returned a complete 1024x1024 image/jpeg (216829 bytes, valid SOI and EOI markers) into result.content.

The call took 15 seconds, which is the concrete case for timeout=180: the buffered path is bounded by call_runner's 5s default, so without it this exact request fails.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.