Extend special casing of sampled_from in unique collections

[DRMacIver]

As of #2030 we have a special case for lists(sampled_from(...), unique=True) that uses a custom strategy. Unfortunately this special casing is trivial to defeat. e.g. lists(sampled_from(...).map(lambda x: x), unique=True) should behave the same way but doesn't. It would be good to extend this support to include arbitrary chains of maps and filters over sampled_from.

There are a couple of possible ways to do this, but I think the easiest would be to extend the base SampledFrom strategy to have a transform function which takes an element of the sample and either transforms it or raises an UnsatisfiedAssumption (or returns a special value). The special case unique list strategy could then call this transformed function directly.

[dfarrel1]

Going to take a stab at this one!

[Zac-HD]

Untested prototype: master...Zac-HD:transformed-samples

[Zac-HD]

We probably also want to add a special case for the st.dictionaries() strategy, which currently doesn't get any advantages at all if keys=st.sampled_from(...). That's now #2687.