ws pool: use ws queue in as_runner

[c-cube]

see if it helps.

[gasche]

Yes, this PR seems to help. I measure like this:

# on main
$ /bin/time timeout 6 sh bench_fib.sh -psize 4 -n 40 -cutoff 8 -kind pool -await
compute fib 40 using await with pool size=Some 4
git checkout improve-ws-poolCommand exited with non-zero status 124
12.57user 5.00system 0:06.33elapsed 277%CPU (0avgtext+0avgdata 2589972maxresident)k
0inputs+16outputs (0major+652285minor)pagefaults 0swaps

# on the branch
$ /bin/time timeout 6 sh bench_fib.sh -psize 4 -n 40 -cutoff 8 -kind pool -await
compute fib 40 using await with pool size=Some 4
fib 40 = 165580141
compute fib 40 using await with pool size=Some 4
fib 40 = 165580141
compute fib 40 using await with pool size=Some 4
fib 40 = 165580141
18.33user 0.36system 0:05.09elapsed 366%CPU (0avgtext+0avgdata 41036maxresident)k
0inputs+10352outputs (0major+40478minor)pagefaults 0swaps

On main, 2589972maxresident suggests that 2GiB of memory were used in 6 seconds. On the branch, 41036maxresident suggests 41MiB.

[gasche]

I think that this is good and that you should merge it :-)