Add __int__, __float__, and __bool__ to the array object

[asmeurer]

These are only required for dimension-0 arrays of a matching dtype.

See numpy/numpy#10404 for why it is not a good idea to require this for higher dimension size 1 arrays.

[oleksandr-pavlyk]

What should the behavior be if the array does not meet the requirement (no zero-dimensional, or different type)?.

Does the standard care whether ValueError or TypeError are raised?

NumPy can raise both:

In [2]: np.__version__
Out[2]: '1.20.0'

In [3]: X = np.zeros((2,2),dtype='d')

In [4]: X.__bool__
Out[4]: <method-wrapper '__bool__' of numpy.ndarray object at 0x2b67bdea50f0>

In [5]: X.__bool__()
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-5-2acecd4ea032> in <module>
----> 1 X.__bool__()

ValueError: The truth value of an array with more than one element is ambiguous. Use a.any() or a.all()

In [6]: X.__float__()
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-6-78817c2ae92b> in <module>
----> 1 X.__float__()

TypeError: only size-1 arrays can be converted to Python scalars

[rgommers]

Does the standard care whether ValueError or TypeError are raised?

In general we've refrained from specifying this, because (a) error handling is broad, (b) existing libraries are very inconsistent, and (c) there are situations where error handling strategy for accelerators may be different from that on CPU on purpose.

[rgommers]

LGTM. Fixed a merge conflict, in it goes. Thanks @asmeurer, and thanks @kgryte and @oleksandr-pavlyk for reviewing.