fix: preserve HTTPX response streams during instrumentation

[shreyas70]

Summary

HTTPX response hooks previously drained the transport response stream to capture its body, then replaced HTTPX's private `_httpcore_stream` state with replay data. That buffered streaming responses, depended on private internals, and could leave streams partially consumed when capture failed.

This change removes that unsafe response-stream consumption and private mutation for all synchronous and asynchronous HTTPX responses.

Behavior change

• HTTPX spans no longer capture `http.response.body`.
• HTTPX response headers and status remain captured.
• HTTPX request headers and body capture remain unchanged.
• Response hooks pass `None` to `generic_response_handler` and never iterate or replace response streams.
• Obsolete response draining, decoding, and replay helpers are removed.

This removes unsafe stream handling implicated by instrumentation. It does not, by itself, prove the production gzip issue's complete causality.

Verification

• Focused TDD regression tests: observed 2 expected failures before implementation, then 2 passed.
• Complete HTTPX suite: 5 passed, 2 existing deprecation warnings.
• Non-database unit suite: 185 passed, 1 skipped, 24 warnings.
• `python -m compileall -q src/harness_sdk`: passed.
• `pylint src/harness_sdk --disable=C,R --ignore-patterns=config_pb2.py`: 10.00/10.
• `git diff --check origin/main...HEAD`: passed.

Database integration tests were excluded because they require the repository's Docker services.

Made with Cursor

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}