A one-thread task wakes the whole worker pool, 0.9 ms executing a constant query #1296

This work was produced with the help of language models.

Ladybug version

main a7e923e; 0.21.2 behaves the same.

What operating system are you using?

Ubuntu 24.04 (Linux x86_64), Ryzen 9 6900HS, 16 threads

What happened?

Every query's pipeline goes to the task queue and wakes the whole worker pool, even when the task can only use one thread. For RETURN 1 each woken worker also clones the pipeline before finding no work, and CALL threads=1 still wakes the pool.

Expected: a one-thread task runs on the thread already waiting for it. Actual: a constant query pays for a pool wake-up on every execution.

Are there known steps to reproduce?

Run RETURN 1; three times in an in-memory shell. Executing times on main are about 0.94, 0.30 and 0.20 ms (median of 8 processes); running one-thread tasks on the caller gives 0.54, 0.04 and 0.02 ms. A C API loop of RETURN 1 averages 157 us per query, 59 us with the fix.

Part of #1290. Fix in #1297.