Generate unique fallback tool-call IDs

[1jehuang]

Summary

• detect provider-synthesized positional tool-call IDs such as bash:0
• replace only those fallback IDs with unique toolu_ IDs
• preserve genuine provider-supplied IDs
• add regression coverage proving uniqueness across responses

Verification

• cargo test -p jcode-provider-openrouter --lib (27 passed)
• GitHub CI: macOS, Windows, Windows cross-target, PowerShell, TypeScript SDK, setup, release, linked-issue, and Greptile checks pass
• Linux is blocked by the pre-existing pinned_todos_payload_stays_empty_when_config_off failure tracked in CI failure: pinned_todos_payload_stays_empty_when_config_off (stale config cache leak from sibling test) #877; Format/Quality are blocked by pre-existing desktop formatting on master

Fixes #884

---

— Jcode agent (automated triage), on behalf of @1jehuang

[greptile-apps]

Greptile Summary

This change prevents positional fallback tool-call IDs such as bash:0 from being reused across independent OpenRouter responses by replacing them with generated IDs. A streamed SSE regression check confirmed that fallback IDs are distinct across responses, while ordinary provider IDs, tool-call flushing, and the final stop reason remain intact.

Confidence Score: 5/5

Safe to merge based on the exercised OpenRouter SSE tool-call flows.

The streamed parser behavior was checked against both the parent and changed implementations: the prior behavior retained a repeated positional fallback ID, while the change generated distinct IDs and preserved normal provider IDs and completion behavior.

Files Needing Attention: No additional files need attention; verification covered crates/jcode-provider-openrouter/src/stream.rs.

T-Rex Logs

What T-Rex did

• Ran the OpenRouter regression test against the parent implementation and observed the fallback ID remained the same (bash:0) for emitted IDs.
• Ran the OpenRouter regression test against the changed implementation and observed all assertions pass: independent streamed SSE responses receive distinct generated fallback IDs, an ordinary provider ID is preserved, tool calls flush at finish_reason: tool_calls, and the message end retains that stop reason.
• Compared pre- and post-change captures and confirmed the parent behavior produced the old ID pattern before capture, while the changed implementation satisfied all assertions after capture; no product code was modified and the temporary test was removed after execution.

_{Ran code and verified through T-Rex}

_{Reviews (1): Last reviewed commit: "fix: generate unique fallback tool call ..." | Re-trigger Greptile}