fix: time out stalled streams

[victor0602]

What changed

• Apply a timeout while waiting for streaming response headers.
• Add a resettable idle timeout between response body chunks.
• Abort and cancel the underlying reader when a stream stalls or its consumer cancels.
• Allow active streams to continue beyond a single timeout interval.

Why

Streaming requests previously removed the timeout signal entirely and could hang forever when a server stopped sending data.

Impact

Stalled streams now fail predictably and release their resources, while healthy long-running streams continue normally.

Checks

• bun test — 453 passed
• bun run typecheck
• bun run lint — no errors; one pre-existing test warning
• git diff --check

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

[NianJiuZst]

No blocking findings on the current head. The header timeout, resettable body-idle timeout, and consumer-cancellation propagation are internally consistent, and I recommend merging this PR on its present base.

Verified locally: typecheck, lint (one pre-existing warning), build, 22 focused HTTP/stream tests, and the full suite (453/453). This is source/mock proof; I did not run a live long-lived provider stream. If another src/client/http.ts PR lands first, this should be rebased and the combined behavior re-tested.