add status reason to UnsatisfiedAssumption rejections

[tybug]

closes #3827.

Note that what we originally discussed in the issue has the knock-on effect of adding to the events of that test case, which shows up in e.g. statistics mode reporting:

from hypothesis import *
from hypothesis.strategies import *

@given(integers())
def test_f(n):
    assume(n > 2)

sandbox.py::test_f:

  - during reuse phase (0.00 seconds):
    - Typical runtimes: < 1ms, of which < 1ms in data generation
    - 1 passing examples, 0 failing examples, 0 invalid examples

  - during generate phase (0.04 seconds):
    - Typical runtimes: < 1ms, of which < 1ms in data generation
    - 99 passing examples, 0 failing examples, 96 invalid examples
    - Events:
      * 49.23%, invalid because: failed to satisfy assume()

  - Stopped because settings.max_examples=100

I'd call this desirable, though perhaps a tad confusing as it doesn't state which assume statement was unsatisfied. But I wonder if there was a good reason this hasn't been done yet.

cc @hgoldstein95

[Zac-HD]

CI is being nitpicky, but this looks great to me - merge once it's passing!

[Zac-HD]

Apparently we have a test for whether we have a test for all error classes with custom init methods!

[tybug]

test_exceptions_are_picklable actually caught a bug due to this, which is that UnsatisfiedAssumption was no longer picklable after making reason a kwarg. To be honest, reason seems sensible as a posarg instead, so I turned it into one (which consequently fixed this issue).

I'm not particularly familiar with pickling issues like this, so maybe there is a more obvious solution.

[tybug]

Ouch. inspect.stack can get expensive, as seen by 91.48s call hypothesis-python/tests/cover/test_searchstrategy.py::test_example_raises_unsatisfiable_when_too_filtered.

from hypothesis import *
from hypothesis.strategies import *

@given(integers().filter(lambda x: False))
@settings(suppress_health_check=list(HealthCheck))
def f(n):
    pass

try:
    f()
except Exception:
    pass

[tybug]

Switched to inspect.currentframe().f_back.f_code.co_name which has a negligible impact.

[tybug]

I would hit the merge button here (you did tell me to!), but I changed some internals since, so I'm going to hold off. Free to yell at me if you think that was a bad call 🙂

[Zac-HD]

That's all about trusting your judgement, and if you want another review I'm happy to do so. In this case it all looks good, so merging now!