Bears-R-Us
3130 numeric test slight revamp
This does several things.
Fixes all the numeric tests that were failing.
Revamps trigonometric and hyperbolic tests, in the same fashion as the _trig_test_helper did (which is now _trig_and_hyp_test_helper). Adds an _infinity_edge_case_helper.
The only trig or hyp function not included is arctan2, since it needs two inputs (numerator and denominator). Adds a denominator array with no zeroes for these cases. There are already zero denominator tests as special cases at the end of test_arctan2, so zero denominators aren't needed in the more general tests.
Includes seed in every test that uses random numbers, so that results will be reproducible.
Combines log and exp tests into one function, because they were identical.
Combines cumsum and cumprod for the same reason.
Sets "ignore" on overflow, invalid data, and divide-by-zero in tests where that info is accurate but unhelpful. This eliminates dozens of runtime warnings that clutter up the test output. This ignoring can be undone by commenting out every line that invokes np.seterr (3 each in _trig_and_test_helper, _infinity_edge_case_helper).
