Investigate GC on long runs

[a2ray]

upto 8,320 cpus for 2-5 hours works ok, but less workers for more time may still have GC issues.

Look at

• https://discourse.julialang.org/t/garbage-collection-not-aggressive-enough-on-slurm-cluster
• or lazy addprocs in Memory leak for lazy worker to worker connections JuliaLang/julia#28887
• https://itensor.github.io/ITensors.jl/dev/faq/HPC.html
• heap_size_hint with -exeflags on workers Interaction between addprocs and --heap-size-hint JuliaLang/julia#50673

[a2ray]

And also,

• it gets worse with some v1.11 releases of julia see here on discourse.
• backport fix at gc: improve mallocarrays locality JuliaLang/julia#56801