Fix ValueErrors raised when numpy iterables use == in failed

[jamescooke]

Adds

• FakeNumpyArray which emulates enough behaviour at test time to have all(result == 1) produce False and abs(left + right) < 19999 raise a ValueError. I couldn't find an easy way to do this with unittest.mock - but would happily see more elegant solutions. Have built out full tests on this fake.
• Handler for ValueError which returns from pytest_assertrepr_compare() when a ValueError is raised - this allows numpy / pandas iterables to fall-back to default pytest output. Fixes ValueError raised when comparing numpy / pandas iterables #40 👈🏻 Some design decisions here.

[hjwp]

wow james, this is awesome. particularly enjoyed the FakeNumpyArray :-D

But i think probably even that is too much to carry around (even if it is lighter than a dependency on numpy).

What do you think about just having numpy as an optional dependency of the tests? We can install it for CI I reckon, and then just do a pytest skip if we find it's not installed on any given developer's machine.

maybe just a try/except around "import numpy" would do it. there's an example here https://stackoverflow.com/a/38442948/366221

[jamescooke]

@hjwp How's that looking now?

[hjwp]

let's do it!

[hjwp]

released as v 0.8.

thanks so much!