Use Numpy API macros as much as possible about bson-numpy

[aherlihy]

The main goal of this ticket was to reduce the amount of code and rely more on the Numpy C API instead of manually manipulating pointers and using memcpy. The bottom line is because we are using named dtypes, a lot of the macros will not work.

PyArray_SetItem: requires a PyObject which would introduce overhead for converting BSON objects into PyObjects only so we could use this instead of memcpy. Additionally, we need to separate out the validation depending on type, so even if we use this instead of memcpy we could not get rid of separate loading functions. Lastly, flexible dtypes that may be shorter than the assigned size (i.e. string, bytes, etc) need to be zero-padded. Could swap out memcpy in simple scalar cases, but would not end up making the code any less complex.

PyArray_GetPtr: See #30

NpyIter_*: We iterate through the BSON object instead of through the ndarray. Can revisit this while implementing #5.

Ultimately -- the issues that we have with the major Numpy C API Macros are not ones that are going away anytime soon, so I believe we can close this issue.

[prashantmital]

It is likely that the Numpy macros now support dtypes since it has had a while to mature. We should reinvestigate doing this.

[prashantmital]

Closing this as we have discontinued development of BSON-NumPy. PyMongoArrow is now the recommended way to materialize MongoDB query results as NumPy ndarrays as well as tabular formats like Pandas' DataFrames and PyArrow Tables.