build: Use build-system pybind11 over Git submodule

[matthewfeickert]

Resolves #309

• Update lower bound on build-system pybind11 to v2.12.0 to add support for compiling for NumPy 2.0.
  • c.f. https://github.com/pybind/pybind11/releases/tag/v2.12.0
• Remove pybind11 submodule to use build-system instead.
• Add an easy to parse log of the environment used for each test, which is useful for debugging differences between CI jobs seperated in time.

For reference, the pybind11 Git submodule was added in 0f9f8ff by @jpivarski, though as this wasn't part of a PR I'm not sure why the Git submodule approach was used over the other pybind11 install methods.

[henryiii]

Why do you have pybind11 and numpy as build requirements? Pybind11 doesn't require numpy at build time. That's one of it's key selling points, actually.

[henryiii]

I don't see any usage of numpy.h.

[matthewfeickert]

Why do you have pybind11 and numpy as build requirements?

Error on my part as I didn't do a git grep to check and just followed Issue #309:

... and to also compile against numpy 2 when building.

I'll fix this now. Thanks for pointing this out.

[matthewfeickert]

I don't see any usage of numpy.h.
...
Why do you have pybind11 and numpy as build requirements? Pybind11 doesn't require numpy at build time. That's one of it's key selling points, actually.

Ah yes, as you point out

fastjet/src/_ext.cpp

Line 20 in 28ec47c

#include <pybind11/numpy.h>

pybind11 takes care of that for us.

[matthewfeickert]

I guess this isn't really necessary(?) to bump it given that there's already a lower bound here, but I guess signaling this also doesn't hurt.

[matthewfeickert]

@lgray can you comment on the original issues that caused you to open Issue #309? I'm not actually clear on what about fastjet should be incompatible with NumPy 2.0.

[matthewfeickert]

As no one has commented yet, I'm going to merge this myself. As always, PRs approved by a single dev can be reverted as needed by the rest of the dev team.