core install: pass sem into every pip call

[bertsky]

No description provided.

[stweil]

This conflicts with #299.

[bertsky]

This conflicts with #299.

I don't care TBH – we are still trying to solve the deployment misery (see CI failures for CUDA variant and recent -j switching), which is urgent.

This is necessary to get a true separation of pip calls between (possibly parallel) venvs.

[kba]

This conflicts with #299.

I think it's a counterargument to #299. And it seems to fix the issue at hand (race condition while installing core). Let's merge and hope for the best. If CD still fails, I will push the images manually.

[kba]

This conflicts with #299.

I think it's a counterargument to #299.

Wrong, sry. We still don't need $(PIP) in the ocrd_all Makefile, only in dependent packages.

[stweil]

I fixed the merge conflict in #299 (now for the second time), so it is ready for merging again.