reproducible builds often fail

Jobs for reproducible builds such as #192100 quiet often fail lately. Sometime it's successful when running on runners with very large /dev/shm

It looks like the increase in /dev/shm space requested in tpo/tpa/gitlab#110 (closed) is no longer enough. Back then we requested around 384MiB of shm, right now it looks like we actually need 451.1MiB.

Options are:

• to ask TPA to make shm even bigger
• to be more selective on what we put there (we actually need the source of 2, maybe 3 dependencies, not every single one)
• finally figure out the actual reason why builds are reproducible when building of tmpfs, but not from other filesystem, and try to fix the underlying issue.
• ?