Slow performance in arrayUtil functions with variable length data

[jreadey]

The functions that deal with converting numpy arrays of variable length elements to a buffer and back can be quite slow.
Running the test program: https://github.com/HDFGroup/hsds/blob/master/tests/perf/arrayperf/bytes_to_array.py with a million element array gave this output:

$ python bytes_to_array.py 
getByteArraySize - elapsed: 0.3334 for 1000000 elements, returned 7888327
arrayToBytes - elapsed: 3.1166 for 1000000 elements
bytesToArray - elapsed: 1.1793

Not surprising since it's iterating over each element in a loop.

Looked into using numba, but numba doesn't work with numpy arrays of object type.
Cython version of arrayUtil?