Decide numpy version based on platform

[sibocw]

1. The latest versions of NumPy do not support macOS with Apple Silicon chips (see torch=2.2.2 is the max supported version for intel macs which is incompatible with numpy>=2 #231).
2. Even with NumPy 2.2.2, I'm having problems with Apple silicon Macs: see CI run at https://github.com/NeLy-EPFL/flygym/actions/runs/13161898909/job/36732489897. Note the following strange errors:

RuntimeError: Could not infer dtype of numpy.float32
RuntimeError: Numpy is not available

Therefore, here I tried to conditionally specify using NumPy 1.* for Apple Silicon Macs and NumPy 2.* otherwise. This is similar to @Dominic-DallOsto's commit 376fe72, but now that we have moved from setup.py to pyproject.yaml, we need to redo it in pyproject.yaml.

However, this made the dependency resolution process (poetry lock) EXTREMELY slow (~35 minutes instead of a few seconds). I don't know why. Until we figure it out, I suggest that we just require numpy<2.0 for all platforms. The aim of this PR is therefore to retain what I changed but not to merge it into ongoing v1.2.0 development.

Does this address any currently open issues?

#231