ARROW-1861: [Python] Rework benchmark suite [skip ci]

[pitrou]

This PR focusses on:

• ASV setup fixes
• splitting the benchmark file
• improving the array from/to pylist conversion benchmarks

[pitrou]

Now ready for review.

[cpcloud]

Can you dedent this 4 spaces.

[pitrou]

Okk.

[cpcloud]

I wonder if this should loop over the array with elements already indexed so that we don't have to compute the difference between time_getitem and time_getitem_as_py to see the overhead of as_py. Something like

for el in self._array2:
    el.as_py()

[pitrou]

Then we'll be benchmarking array iteration ;-) Instead I think I'll store the list of indexed values in the setup() method.

[cpcloud]

Fair enough.

[pitrou]

Note we may want to benchmark as_py() on different types.
(though the inner machinery may already be exercised as part of to_pylist?)

[cpcloud]

Probably ok to blow away the commented out things

[cpcloud]

LGTM

[pitrou]

By the way, do you have any ideas for further benchmarks?
For example, is it useful to benchmark creation of a pa.Array from a Numpy array?

[xhochy]

+1, LGTM