test: unpin hypothesis and bump version

[ahal]

No description provided.

[ahal]

The issue that forced the pin was fixed in:
HypothesisWorks/hypothesis#4639

However, the maintainers pointed out that we're setting things up poorly:

Thanks Andrew. To clarify, the issue was the json strategy passes lambda x: useful_list | useful_dict as the extension strategy to st.recursive, but this doesn't actually use x and therefore doesn't recurse more than one level.

We started assuming this in the new min_leaves feature, and have now both relaxed this assumption so this code works again, and added a warning when we detect this class of miss-stated extend.

I recommend adjusting this strategy to recurse on x regardless, as its test distribution is probably weaker than expected at the moment!

I'll attempt to also fix up the test.

[jcristau]

@ahal worth getting this landed, and file a bug/issue to fix the test as a followup?