Rework the scheduler to use threads instead of processes

Description

This removes the usage of processes in the scheduler to use threads instead.

This is the first step towards #1389.

This implements the changes required to run with a threaded scheduler.

Things to test

I have tried extensively to stop/start/kill the processes to ensure that everything still behaves accordingly.

However, more eyes would be good. Please try breaking it!

Profiles

Two profiles to show the difference between master and now:

For source fetch:

• master
• this branch

For build

• master
• this branch

Benchmarks

Those benchmarks use the new forkserver setup for running the blocking jobs.

• Finished in: 95 mins, 15 secs
• Number of builders: [4, 8, 12]
• Target files: base-files/base-files.bst
• Number of runs: 3
• Number of warmups: 1
• Python versions: py38

action	python_version	commit	median time (secs)	mean time (secs) ± std
build - 12	py38	bschubert/no-multiprocessing - 2e592a38	72.18	72.61 ± 1.22
	py38	master - 7fee0e64	98.12	97.82 ± 0.66
build - 4	py38	bschubert/no-multiprocessing - 2e592a38	74.38	75.15 ± 1.83
	py38	master - 7fee0e64	110.59	108.34 ± 5.36
build - 8	py38	bschubert/no-multiprocessing - 2e592a38	72.93	73.54 ± 1.67
	py38	master - 7fee0e64	104.23	105.04 ± 1.93
show	py38	bschubert/no-multiprocessing - 2e592a38	5.24	5.27 ± 0.05
	py38	master - 7fee0e64	5.14	5.35 ± 0.38
show - cached	py38	bschubert/no-multiprocessing - 2e592a38	11.42	11.48 ± 0.20
	py38	master - 7fee0e64	11.25	11.28 ± 0.07
show - sources fetched	py38	bschubert/no-multiprocessing - 2e592a38	8.19	8.10 ± 0.41
	py38	master - 7fee0e64	7.49	7.60 ± 0.32
source fetch	py38	bschubert/no-multiprocessing - 2e592a38	600.05	599.23 ± 1.61
	py38	master - 7fee0e64	87.39	86.86 ± 1.80