fix(libsy): fall back on efficient stream transport errors

[ting-hong-shieh]

What

• Fall back to the capable target when the efficient response stream raises a transport error during aggregation.
• Preserve other aggregation failures as typed client-call errors.
• Add regression tests for the transport fallback and the non-transport error boundary.

Why

The efficient client call completes as soon as it returns a stream handle. A transport error can surface later, while into_agg() consumes that stream for the escalation judge. That error was converted to AlgorithmError, so the route returned HTTP 500 without calling the capable target.

This change treats only a mid-stream transport failure as recoverable. Other aggregation errors keep their original client error type instead of silently changing targets.

Closes #388

How tested

• cargo fmt --all --check
• cargo clippy --workspace --all-targets -- -D warnings
• cargo test -p switchyard-libsy
• cargo test --workspace

Checklist

• Unit tests cover the bug fix and its non-transport boundary.
• No public API or customer-facing documentation changed.
• Commit includes the required DCO sign-off.

Notes for reviewers

The fallback is intentionally limited to LlmClientError::Transport; invalid or malformed responses remain structured client-call failures.

Summary by CodeRabbit

• Bug Fixes
  • Improved escalation routing reliability by falling back to a capable target when streamed responses encounter transport failures.
  • Preserved specific client errors for other response aggregation failures.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 7cdd54d7-c9b5-4c43-83da-e066792121ce

📥 Commits

Reviewing files that changed from the base of the PR and between 48b3b71 and bab53cb.

📒 Files selected for processing (1)

• crates/libsy/src/algorithms/llm_class.rs

---

Walkthrough

The escalation classifier now falls back to the capable target after efficient-stream transport failures. Other aggregation failures remain typed ClientCall errors. Tests cover both streamed failure paths.

Changes

Escalation stream fallback

Layer / File(s)	Summary
Aggregation error routing crates/libsy/src/algorithms/llm_class.rs	Efficient response aggregation now routes transport failures to the capable target and converts other aggregation failures into typed client-call errors.
Stream error regression coverage crates/libsy/src/algorithms/llm_class.rs	Tests cover partial streamed responses followed by transport or non-transport client errors. They verify capable-target fallback and efficient-model error preservation.

Estimated code review effort: 3 (Moderate) | ~20 minutes

Poem

I’m a rabbit with a stream in my ear,
Transport thumps no longer cause fear.
The capable target hops into place,
Typed errors keep their proper trace.
Tests cheer, “The route is clear!” 🐇

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly identifies the transport-error fallback fix in the escalation routing changes.
Linked Issues check	✅ Passed	The changes address transport-error fallback and preserve non-transport aggregation errors, with regression tests covering issue #388.
Out of Scope Changes check	✅ Passed	The changes are limited to escalation aggregation error handling and related regression tests described in issue #388.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.

---

_{Comment @coderabbitai help to get the list of available commands.}

[ayushag-nv]

Nice Catch. Thanks for the contribution. Approving