refactor: Improve makefile build commands

[orlp]

This does a couple things:

1. We now properly match the CPU flags to that of our release. Specify LTS_CPU=1 e.g. make build-release LTS_CPU=1 to build a lts-cpu variant.
2. You can now pass along arguments to maturin through the makefile commands, e.g. make build ARGS="--no-default-features".
3. I redefined the build profiles. We now have (ordered by compilation speed):

Name	Optimization	Debug assertions	Debug symbols	Compilation Speed
debug	No	Yes	Yes	Fastest
nodebug-release	Yes	No	No	Slow
release	Yes	No	Minimal	Slower
debug-release	Yes	No	Yes	Slower still
dist-release	Extreme	No	No	Slowest

Almost all the time you want debug, which is just make build. If you are investigating performance, you want make build-release.

The remaining profiles are more niche. Use dist-release to exactly match the optimization we use when distributing, which is not necessary most of the time. The nodebug/debug variants of release exist because I found it adds significant (30-60 sec) compile time between them (all fresh compiles):

Apple M2
make build 1:50
make nodebug-release 4:51
make release 5:33
make debug-release 6:39
make dist-release ~20 (couldn't be bothered to run it)

Use nodebug if you are iterating on something performance-sensitive but you've already done profiling so you know what to improve without needing further profiling. Use debug when hunting down a bug with a debugger that only occurs in release mode.

[codspeed-hq]

CodSpeed Performance Report

Merging #19498 will improve performances by 35.4%

_{Comparing orlp:makefile-build (64ad3f5) with main (6a7d140)}

Summary

⚡ 1 improvements
✅ 40 untouched benchmarks

Benchmarks breakdown

	Benchmark	main	orlp:makefile-build	Change
⚡	test_to_numpy_series_with_nulls	435.7 µs	321.8 µs	+35.4%