Fix compile_targets.py when using numpy 1.24.0 or newer

[pekkarr]

Fixes #217

Numpy 1.24.0 removed creation of arrays with mismatched dimensions to have implicit dtype=object. To maintain the old behavior, dtype=object has to be specified explicitly.

https://numpy.org/neps/nep-0034-infer-dtype-is-object.html

This changes the return type of LicenseInfo.toNumpyString() to a numpy array. The method is used in three places:

• compile_targets.py: Previously the tuple return value was converted to a numpy array implicitly by the np.save method. With this change, it gets a numpy array instead of a tuple, so the behavior is exactly the same as with older numpy versions.
• core/algos3d.py: The return value is passed to np.ascontiguousarray, which also converts arguments implicitly to numpy arrays, so this should work with the change as well.
• shared/proxy.py: The return value is destructured to two variables. Numpy arrays support this so it isn't affcected by the change.

[joepal1976]

So far validated not causing problems on earlier numpy (1.17.4)

[joepal1976]

Can't see any obvious downside of this. Going to merge so it ends up in nightly and see if we feedback from somewhere else.