Force push on a release branch causes race conditions due to concurrent pipelines

[alalazo]

If a release branch in Spack is force-pushed, when a pipeline for it is still running, that currently triggers a second pipeline that is not synchronized with the first one. Example:

• (First run triggered by pushing the releases/v0.20 branch) https://gitlab.spack.io/spack/spack/-/pipelines/376508
• (Second run triggered by removing a commit from the branch above) https://gitlab.spack.io/spack/spack/-/pipelines/376533

This caused a bunch of sha256sum errors due to race conditions among pipelines. Canceling the first pipeline, for some reason, didn't cancel all the jobs.

Additional notes

If two pipelines are running and pushing concurrently to the same buildcache, that gives us no guarantee on the final state of the buildcache¹

Footnotes

1. For instance, the second pipeline might be ✔️ but the buildcache be inconsistent and have some packages pushed ultimately by the first pipeline. ↩