agent: retry transient LLM failures in loop

[pikaxinge]

Summary

• retry transient provider failures in the agent loop using existing error classification
• keep context-window retry/compression behavior unchanged for context exhaustion
• add explicit retry notice to non-internal channels during transient retries
• add tests for transient retry success and non-retryable error behavior

Test Plan

• go test ./pkg/agent -run 'TestAgentLoop_TransientLLMErrorRetry|TestAgentLoop_NonRetryableLLMError_NoRetry|TestAgentLoop_ContextExhaustionRetry'
• go test ./pkg/...

Fixes #629

[Copilot]

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Zhaoyikaiii]

therre have two kinds retry?

• time out、502/503、connection reset、TLS -> backoff retry ?
• contains(token/context/invalidparameter/length) -> compression retry?

[nikolasdehor]

This PR is superseded by #866, which extracts the retry logic into a shared pkg/utils/llm_retry.go engine and addresses the reviewer feedback from #594 (parent deadline capping, Retry-After parsing, jittered backoff, toolloop integration).

The core changes here (transient retry in agent loop, isTransientLLMError, isContextWindowError) are all present in #866 in a more complete form.

Recommend closing this in favor of #866.