solidcraft-smlib held one process-wide kernel lock for the whole of every
solidcore call, so a long call in one design (a modeled thread, a fillet)
held up every kernel call in every other design, even after its tab was
closed, and a call that never returned held them until restart.
Each call now takes a kernel from a pool and the lock is only held to take
or put one back. Calls one after another get the same kernel as before; one
that starts while another runs gets its own. Each kernel has its own range
of procedural blend serials, since a body keeps the serials of the kernel
that made it.
History commands and delete checks now belong to their design: they wait
with its tab, go when it closes, and can't land in another design. An
agent's command for a design that left the window is refused at once
instead of holding up the requests behind it.
Closes#11681
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>