Enable parallel builds by default.

[thatch]

This roughly halves the time taken to build on a 12-core MBP to make it easier to iterate. I picked 3.10 arbitrarily but this affects all make invocations by default. One can override it with cpu_count=1 for a particular call if there are problems.

before change

INFO Compiling cpython took 6 minutes 26 seconds
INFO Tar build/cpython-3.10.19/ppp-marker/3.10.19 -> dist/cpython-3.10.19-macos-arm64.tar.gz
INFO Overall compilation took 9 minutes 56 seconds
portable-python build 3.10.19  321.52s user 120.27s system 74% cpu 9:56.96 total

after change

INFO Compiling cpython took 3 minutes 42 seconds
INFO Tar build/cpython-3.10.19/ppp-marker/3.10.19 -> dist/cpython-3.10.19-macos-arm64.tar.gz
INFO Overall compilation took 4 minutes 51 seconds
.venv/bin/portable-python build 3.10.19  327.72s user 141.55s system 160% cpu 4:52.06 total

[thatch]

The more I think about it, -j0 is nonsensical; why don't we make that the "auto" default, and forbid negative numbers entirely? I'm not sure when I'd ever need almost-all-the-cores.

[zsimic]

Nice speed up!

[thatch]

More results, this time GCC on Linux. A big chunk of the time is spent on two link steps, so most of the win comes from being able to do those two concurrently.

	Before	After	Speedup
12-thread Xeon (HT) - Archlinux	29m56s	11m46s	2.54x
12-thread Xeon (HT) - Ubuntu	22m42s	10m6s	2.25X
40-thread Xeon (HT) - Archlinux	31m47s	12m16s	2.59x

[zsimic]

Excellent bump! Let's merge it. We can also change/review this to be "use all CPUs" by default, with just a boolean to turn it off as well... You're right, the granularity on how many CPUs to use is not really needed, it was just mimicking -j in a way. So we could change this from cpu_count=None to parallel=True, I'm down for the simplification if you think it'll help future readers.