Prefer x86_64_v2 Python builds

[j178]

Summary

Resolves #8499

[konstin]

Thank you!

[konstin]

Nit: This can be a method on Arch

[konstin]

I check that the versions uv uses all run on my machine.

CC @zanieb for the changes in the generator script.

[zanieb]

Can we use these unconditionally? Don't we need to check if the system supports the v2 instruction set before selecting these distributions?

[konstin]

According to https://gregoryszorc.com/docs/python-build-standalone/main/running.html:

x86_64_v2-*

Targets 64-bit Intel/AMD CPUs approximately newer than Nehalem (released in 2008).
Binaries will have SSE3, SSE4, and other CPU instructions added after the ~initial x86-64 CPUs were launched in 2003.
Binaries will crash if you attempt to run them on an older CPU not supporting the newer instructions.

Requiring at least a cpu from 2008 is safe enough. I'd like to add cpu feature detection on top of that (e.g. avx2 for x86_64_v3), but this change in itself is a good idea.

[zanieb]

I think we should probably retain the x86_64 downloads (so you can install them by key) then add and prefer the x86_64_v2 downloads by default. Removing access to the x86_64 downloads entirely doesn't make sense to me.

Also, this is a breaking change; although it may not be likely. We might want to roll it into 0.5.0.

[zanieb]

I'm not particularly enthused about defaulting to the v2 builds without checking the CPU features, especially since we need to do so anyway in the future. But I feel less strongly about that as long as there's a path towards recovering the existing behavior.

[danielhollas]

Are there any benchmarks that show whether / how much CPython is faster with these extra instructions? Might by interesting to run the pyperformance test suite (which is what faster CPython team uses for benchmarking).