Add boxing and unboxing functions for dpnp.ndarray to numba_dpex

[diptorupd]

To support dpnp.ndarray as input args to numba-dpex's version of njit decorator, numba-dpex should be able to convert a dpnp.ndarray PyObject to a native Numba representation and vice-versa at function boundaries.

[diptorupd]

@oleksandr-pavlyk I have made some progress on the boxing and unboxing support for dpnp.ndarray. The latest version of the WIP code is in : https://github.com/IntelPython/numba-dpex/blob/dpnp_ndarray_update_v3/numba_dpex/core/types/dpnp_ndarray_type.py.

I am considering switching to dpctl_capi.h to not have to reinvent the getter functions for unboxing a dpnp.ndarray. Before I make any changes here are few questions:

1. The getters defined in _usmarray_api.h should all work on a PyObject of type dpnp.ndarray as they simply do the equivalent of getattr? [EDIT: I think it is trickier then that as these functions expect a PyUSMArrayObject pointer]
2. I do not see any setters or constructors for UsmNDArray in the CAPI. To return a dpnp.ndarray (boxing) from an njit function should I create a PyUSMArrayObject object directly and then set the needed attributes?
3. If 2) above is a correct assumption, how do I use the object to create a dpnp.ndarray from the C API? I was thinking of calling the interpreter and somehow call the dpnp.ndarray constructor and pass in the PyUSMArrayObject and return that PyObject.

Advise will be much appreciated.

[oleksandr-pavlyk]

You are right, there are no constructors in _usmarray_api.h, only getters at the moment, and these expect PyUSMArrayObejct.

At the moment, the recommended way to create usm_ndarray is to use C-API Memory_Make to create memory object mobj from existing USM allocation, and then call Python interpreter to execute usm_ndarray(shape, dtype=dtype, buffer=mobj).

I will add UsmNDArray_Make(int nd, ssize_t *shape, int typenum, c_dpctl.memory._Memory mobj) to dpctl's C-API to eliminate the need to call the interpreter and the need to create shape tuple.

[diptorupd]

Completed in #902