Allow num_regression to accept array-like inputs

[jpn--]

Presently, the num_regression fixture only accepts actual 1-d numpy arrays. It would be convenient if it would also accept array-like inputs, including list of numbers and scalars. It's easy enough to convert them in my tests, but it's also super easy and suitably concise to make the change here. I'm not able to simply use the regular data_regression fixture in cases where the values are floats that might have a tiny bit of variance. The num_regression fixture with its tolerance control is exactly what I want in that instance.

[nicoddemus]

This is looking great already, thanks @jpn-- for the contribution!

Please take a look at my two minor comments. 👍

[tadeu]

Awesome, thanks for this improvement :)

[nicoddemus]

Thanks a lot @jpn--!