BLD: oldest-supported-numpy is now unneeded

[neutrinoceros]

Since numpy 1.25, building backward compatible wheels against newer versions of numpy is supported, see:
https://numpy.org/doc/stable/release/1.25.0-notes.html#compiling-against-the-numpy-c-api-is-now-backwards-compatible-by-default

As a result, oldest-supported-numpy will likely stop being updated in the near future and we should migrate back to numpy>=1.25 as a build-time requirement instead.

I currently have no reasons to think this migration is urgent, but depending on how things play out in the coming weeks it might be necessary to do it in order to add compatibility with Python 3.12

It may still be worth trying it out sooner rather than later, if time allows it.

[neutrinoceros]

important note: numpy 1.25 doesn't support Python 3.8, so oldest-supported-numpy is still needed for that specific version.

[neutrinoceros]

Xref scipy/oldest-supported-numpy#76

[neutrinoceros]

One unexpected lesson from the discussion upstream is that wheels built with numpy 1.x (e.g., using oldest-supported-numpy) will not be runtime-compatible with numpy 2.0 (due early 2024), but the new backward-compatibility mechanism introduced in numpy 1.25 will allow wheels built on 2.0 to run fine with older numpys. With this in mind, migrating away from oldest-supported-numpy appear like a necessary step towards compatibility with numpy 2.0, so let's give it a try.