Streaming HTTP errors raise ResponseNotRead

[saxon134]

GlobalRouter.stream() and stream_async() send requests with stream=True, then pass a >=400 response directly to error_from_response() without reading the body first. For a real streaming response, response.json()/response.text access raises httpx.ResponseNotRead, so callers do not receive the documented GlobalRouterError. Reproduce with a mock SyncByteStream returning a 400 JSON error body and calling client.stream(...); it raises ResponseNotRead. Read/aread the response before normalizing the error.

File: src/globalrouter/_client.py

Line: 214

Severity: medium

Summary: unread streaming response in globalrouter.stream

[saxon134]

Coding Loop Plan

Refined Issue

Streaming requests that receive HTTP error responses currently raise httpx.ResponseNotRead instead of the SDK's normalized GlobalRouterError because the error parser reads JSON/text from a response opened with stream=True before the response body has been consumed.

Implementation Plan

Implementation Plan

Fix only the streaming HTTP-error path. Add regression coverage showing streamed sync and async HTTP error responses produce GlobalRouterError with parsed status/message/code metadata, then update the stream helpers to read error response bodies before passing them to existing error normalization. Keep successful streaming behavior unchanged and avoid unrelated retry or resource refactors.

Development Tasks

1. Add regression tests for streaming HTTP error normalization
   Add tests that use httpx.MockTransport to return a JSON HTTP error response for a streaming endpoint. Cover the sync client.chat.stream(...) path by forcing iteration with list(...), and cover the async client.chat.stream_async(...) path by consuming it in an async loop. Assert the raised exception is GlobalRouterError, not httpx.ResponseNotRead, and assert key normalized fields such as status_code, message, code, error_type, and request_id when metadata is present.
   Files: tests/test_client.py
   Tests: python -m pytest tests/test_client.py -q

2. Read streamed error bodies before error parsing
   Update GlobalRouter.stream so that after send(..., stream=True) returns a response with status_code >= 400, it calls result.read() before error_from_response(result), then closes the response in the existing finally block. Update GlobalRouter.stream_async similarly with await result.aread() before calling error_from_response(result). Do not change success streaming behavior, non-stream request behavior, or retry policy.
   Files: src/globalrouter/_client.py
   Tests: python -m pytest tests/test_client.py -q

3. Run full project validation
   Run the repository's normal validation commands after the focused fix passes. Address only failures caused by this change.
   Files: src/globalrouter/_client.py, tests/test_client.py
   Tests: python -m pytest -q, python -m ruff check ., python -m mypy src/globalrouter tests