[Solaris-star]

Fixes #504.

The LLM queue returns one syscall per getMessage() call. The RR worker passed that object directly to _execute_batch_syscalls(), whose first iteration raises TypeError before reaching the LLM adapter. This still reproduces on main at 292d98f7.

Wrap the queue item in a singleton list at the dispatch boundary. The batch helper and its existing multi-item API stay unchanged.

Tests

Added four cases covering an empty queue, one request, consecutive requests, and direct multi-item batch dispatch. The queue helper, scheduler, base scheduler, and Syscall class are real; the LLM executor is mocked.

• Baseline: 2 failed / 2 passed, with the expected non-iterable-syscall error.
• Fixed: 4 passed; an independent rerun also passed all four.
• Local execution isolated unused backend/context imports and the Query annotation dependency. Ordinary pytest collection in the available environment is blocked by missing cerebrum; the full dependency installation, full suite, and Ollama/Qdrant integration tests were not run.
• Ruff 0.5.0 passes the new test file. The scheduler file has the same 11 lint diagnostics as unchanged main. Compile and whitespace checks pass.

No public API or configuration change, so no user documentation update is needed. Keeping this draft pending broader environment validation and maintainer review.