Stop agent follow-up actions after tool failures

[tsubasakong]

Summary

Fixes #414

• inject explicit no-fabrication guidance into the next LLM turn whenever any tool call fails
• apply the guidance in both standard and streaming agent loops so failed web_search results are not treated as evidence for downstream actions
• add a regression test that asserts the guidance is included after a failed tool call

Changes

• inject explicit no-fabrication guidance into the next LLM turn whenever any tool call fails
• apply the guidance in both standard and streaming agent loops so failed web_search results are not treated as evidence for downstream actions
• add a regression test that asserts the guidance is included after a failed tool call

Testing

• cargo clippy --workspace --all-targets -- -D warnings passes
• cargo test --workspace passes
• Live integration tested (if applicable)

Testing notes:

• git diff --check
• host Rust toolchain unavailable here (cargo/rustc not installed), so I could not run cargo test

Fixes #414.

Security

• No new unsafe code
• No secrets or API keys in diff
• User input validated at boundaries