test(sandbox): cover inference stream truncation SSE errors

[mjamiv]

Summary

• Add live-upstream sandbox proxy tests for all three streaming truncation paths from inference proxy: no end-to-end test coverage for the three truncation → SSE-error paths in route_inference_request #846: byte limit, upstream read error, and chunk idle timeout.
• Exercise the real routing path through process_inference_keepalive -> route_inference_request -> Router::proxy_with_candidates_streaming -> reqwest live body, instead of adding a production mock streaming API.
• Keep production streaming limits unchanged; the smaller test cfg constants only keep unit tests deterministic and fast.

Closes #846.

Validation

• cargo fmt --all -- --check
• cargo test -p openshell-sandbox inference_stream_ -- --nocapture
• cargo test -p openshell-sandbox
• cargo clippy -p openshell-sandbox --all-targets -- -D warnings

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[github-actions]

Label test:e2e applied, but pull-request/1418 is at {"messa while the PR head is c0a7e9f. A maintainer needs to comment /ok to test c0a7e9f0799db527630d6733744c2b436bf067e2 to refresh the mirror. Once the mirror catches up, re-run Branch E2E Checks from the Actions tab.

[TaylorMutch]

/ok to test c0a7e9f

[TaylorMutch]

LGTM, thank you!