Use --fast-ci/--slow-ci default for parallelism

[vstinner]

No longer pass a pessimistic -j2 option by default. Instead, rely on --fast-ci/--slow-ci which uses -j0 by default: use os.process_cpu_count() or os.cpu_count() to decide the number of jobs to run.

[vstinner]

--fast-ci and --slow-ci uses -j0 by default: https://github.com/python/cpython/blob/main/Lib/test/libregrtest/cmdline.py#L480-L483

When -j0 is used, libregrtest uses (os.process_cpu_count() or 1) + 2: https://github.com/python/cpython/blob/565f76fee1fe1f3ea39aee203e2ed4887a67bc11/Lib/test/libregrtest/main.py#L539-L541

os.process_cpu_count() was added to Python 3.13 and is different than os.cpu_count(): it takes in account CPU affinity: https://docs.python.org/dev/library/os.html#os.process_cpu_count

Example:

# Run Python on two specific CPUs using taskset
$ taskset -c 5,11 python -c 'import os; print(os.cpu_count())'
12
$ taskset -c 5,11 python -c 'import os; print(os.process_cpu_count())'
2

[StanFromIreland]

This will reduce parallelism for UnixInstalledBuild which invokes regrtest directly with no --slow-ci.

Also, it may be a little risky for sanitisers to have no cap...