Streaming HTTP errors are parsed before the body is read

[saxon134]

GlobalRouter.stream() calls error_from_response() on the response returned by httpx.Client.send(..., stream=True) without first reading the response body. For a real streaming response, error_from_response() reaches response.json()/response.text before read() has been called, causing httpx.ResponseNotRead instead of the documented GlobalRouterError for a 4xx/5xx streaming endpoint. The async path has the same issue and should await result.aread() before normalization.

File: src/globalrouter/_client.py

Line: 214

Severity: medium

Summary: unread stream response in stream

[saxon134]

Coding Loop Plan

Refined Issue

Streaming HTTP error responses are converted to GlobalRouterError before their streamed body has been read, so error_from_response may raise httpx.ResponseNotRead instead of parsing the server's error payload.

Implementation Plan

Implementation Plan

Add focused regression coverage for sync and async streaming HTTP error responses, then update the streaming error paths to read the response body before normalizing the error. Keep the change limited to the client streaming methods and their tests.

Development Tasks

1. Add streaming HTTP error regression tests
   Add tests that simulate a non-2xx response returned from a streamed request with a JSON error body. Cover both GlobalRouter.stream through a public sync streaming method, such as client.chat.stream(...), and GlobalRouter.stream_async through the matching async method. Assert that GlobalRouterError is raised, not httpx.ResponseNotRead, and that status_code, message, code/error_type/request_id when present are parsed from the JSON body.
   Files: tests/test_client.py
   Tests: python -m pytest tests/test_client.py::test_sync_streaming_http_error_reads_body_before_parsing tests/test_client.py::test_async_streaming_http_error_reads_body_before_parsing -q

2. Read streamed error bodies before parsing
   In GlobalRouter.stream, when result.status_code >= 400, call result.read() before passing the response to error_from_response(result), and still close the response in the existing finally block. In GlobalRouter.stream_async, do the async equivalent with await result.aread() before error_from_response(result), and keep await result.aclose() in the finally block. Do not change successful streaming behavior or retry behavior.
   Files: src/globalrouter/_client.py
   Tests: python -m pytest tests/test_client.py::test_sync_streaming_http_error_reads_body_before_parsing tests/test_client.py::test_async_streaming_http_error_reads_body_before_parsing -q, python -m pytest tests/test_client.py -q