ENH, TEST: silence prints when testing -m full

[mattip]

Closes #20289

• stop printing spurious values in an f2py test
• when calling np.empty([200, 200, 0]), only allocate the minimum data required (a single element). Previously this would allocate 200 * 200 * 1 elements.

[seberg]

IIRC, there was some weird hesitance about doing this change? If the only hesitance was the "undefined behaviour" of working with pointers outside of the allocated range (but never de-referencing them), I am willing happy to give this a shot if everyone thinks that this in practice just should not matter.

@eric-wieser do you remember if it is important to avoid that "undefined behaviour"? I.e. in your PR here: gh-15788 you tried to ensure no "out-of-bound" pointer arithmetic is possible by overriding strides to 0?

[mattip]

Closing, I forgot about #15788 which is more complete