Incorrect behavior of order parameter for order="K"

[ndgrigorian]

As has been picked up by the array API test suite, the order="K" parameter does not work in some cases:

In [1]: import dpctl.tensor._usmarray as ua, dpctl.tensor as dpt

In [2]: x = dpt.usm_ndarray((1, 1, 2, 1, 1), dpt.bool, (0, 0, 1, 0, 0))

In [3]: dpt.astype(x, dpt.bool, order="K")

this raises
ValueError: buffer='<SYCL(TM) USM-device allocation of 2 bytes at 0xffffb80200010200>' can not accomodate the requested array.
because the logic calculating the strides is incorrect in this case.

dpt.copy raises the same error for the same array

The internal function _empty_like_orderK does better

In [3]: x = dpt.usm_ndarray((1, 1, 2, 1, 1), dpt.bool, (0, 0, 1, 0, 0))

In [4]: _empty_like_orderK(x, dpt.bool)
Out[4]:
usm_ndarray([[[[[False]],

               [[False]]]]])

but it's worth noting that the output of the strides are (1, 1, 1, 1, 1) where for Numpy they are (2, 2, 1, 1, 1).

This behavior should be implemented identically everywhere order="K" appears in dpctl.

[oleksandr-pavlyk]

This is being fixed in gh-1324

[ndgrigorian]

Resolved by #1324