feat: add pipes supporting for numpy objects in the pipeline

[hollandjg]

Description

Provide some new pipe-functions (Pipeline steps) which convert pool sequences into arrays.
Sequences it accept look like range(5), zip(range(0, 10), range(0, 10)), and the functions pass a numpy array or a numpy recarray to the next pipeline step.

The general idea is that you drop the pipe sequence_to_array into your pipeline after a step which uses one of the more primitive pooling functions, and it'll make you a numpy array of the right type so you don't have to worry about it in the next pipeline step. You can switch the results back to the sequence form at the end, by using array_to_sequence.
Something like

Pipeline([
    ("pool", grid_pool),
    ("convert_for_next_steps", sequence_to_array),
    ("summed_dissimilarity_sampler", summed_dissimilarity_sampler),
    ("revert_to_original_form_for_output", array_to_sequence),
])

... and we could then take out the complex coercing of the input from the summed_dissimilarity_sampler, say.

Type of change:

• New feature (non-breaking change which adds functionality)

[benwandrew]

nifty! works like a charm. wonderful documentation, and i got to learn about recarray!