Add guide to designing strategies for good shrinking behaviour.

[Zac-HD]

Closes #1705.

I'd appreciate lots of feedback on this as I certainly don't think it's perfect - but hope it's better than nothing.

[DRMacIver]

Thanks for doing this. It generally looks very good!

One thing that is missing that I think is worth including as the most common problem is the use of length parameters: Things like integers(0, 10).flatmap(lambda n: st.lists(blah, min_size=n, max_size=n)). They work in the sense that Hypothesis makes certain guarantees about being able to shrink them, but they result in shrinking being much slower than it otherwise would be.

[DRMacIver]

I'd actually forgotten this trick. It's a good trick.

[Zac-HD]

One thing that is missing that I think is worth including as the most common problem is the use of length parameters: Things like integers(0, 10).flatmap(lambda n: st.lists(blah, min_size=n, max_size=n)).

That's covered under "keep things local", though the only examples there are in the @composite / inline-draw style. I'll add an example of each in the method-chaining style and comment that they're equivalent under the hood.

[Zac-HD]

(misclick...)

[Zac-HD]

CC @DRMacIver - I've overhauled the last section on internal APIs Anything else you think needs to be handled before merging?

[DRMacIver]

I haven't actually done more than a brief skim of a rereview, but based on the fact that a) I'm not likely to this week due to other considerations and b) This is already much better than the status quo of not having it, I figured we might as well merge. Thanks for writing it!