update requirement to allow numpy>2

[CarlosCancino-Chacon]

The current requirements specify a numpy version <2.0. However, some packages are starting to require numpy >2. This is a simple update that updates the requirements.

TODO:

• Check that this update doesn't break anything in the other related packages (the benchmark and the demo).

[darth-alexus]

I also checked it on my machine: nothing breaks, tests pass on both numpy 1.26 and 2.5 :)

One thing to mention: I tested it on the Bach fugue example: numpy 2 changes some alignment outputs slightly. 4 of the 15 methods give different warping paths (audio/arzt, audio/dixon, midi/SL_OLTW, plus pure float noise in audio/skf). Only a handful of frames, at most ~0.4 beats off. The results are deterministic, and the differences stay when scipy/librosa are held at the same versions, so this should come from numpy itself.

For MIDI: the cause is numpy 2's default sort ordering ties differently, so partitura hands back simultaneous (chord) notes in a different order. The audio cases are probably the same thing, since the score gets synthesized from that note array, but I didn't check that.

The benchmark CI installs pymatchmaker@develop without pinning numpy atm, so once this merges it'll quietly switch to numpy 2, while environment.yml still pins 1.26.4. And the references computed before and after won't match exactly. But this shouldn't be an issue because we will compute the references again anyways.

But I'd suggest we pin the same numpy in the benchmark workflows and in environment.yml, so that we won't have any differences or (un)pleasant surprises for new submissions .