[Crivella]

Ugly but working fix for issue observed in:

• {2025.06}[foss/2025b] AITW-wood_stiffness_homogenization, AITW-aiida-wood-permeability software-layer#1871 (comment)

Should also work for:

• {2025.06}[GCCcore/14.3.0] modkit v0.6.4 software-layer#1543
• https://gitlab.com/eessi/support/-/work_items/270

if the failures are truly random and just rerunning the build multiple times will cause it to succeed.
In this case we leverage the fact that re-running the pip install command will skip already installed crates and just process the not yet compiled ones

Manually tested by running it 3 times successfully on top of EESSI/software-layer#1871
The last iteration with the print of the failures failed 3 times before succeeding.

TODO

• Cleanup comments (i've left them there for now to show some of the things that have been tried)
• Potentially properly find the source of the failure but i think it is something very specific to our nodes + Rust/LLVM version combination. This failure does not seem to occur in our 2026.1 builds and i assume was never observed in the 2023.06 ones, or on other haswell machines with the same software stack