Check frule primal with isapprox

[sethaxen]

frule_test currently requires the primal output of frule of f to be exactly equal to the output of f. We previously relaxed this requirement for rrule_test (#33), and this PR just uses the same check for frule_test.

The ChainRules test suite still passes for me with this change.

[willtebbutt]

Hmm my concern is that this is going to break when the functions output objects for which isapprox is not defined, such as Tuples. I don't think that we test enough things in this package to be sure that this won't break lots of things downstream unfortunately. Possibly we need to implement our own function to check approximate equality. @oxinabox any thoughts?

[sethaxen]

Yes, rrule_test has this same limitation. One easy thing we could do for both frule_test and rrule_test would be conduct a strict equality check first, followed by an approximate equality check if that fails. This would ensure that this test passes for everything it currently passes for.

[willtebbutt]

Hmm yeah, I would definitely be on board with that as a short- to medium-term solution.

[sethaxen]

Okay, I added back the equality check first, and I did the same thing for rrule_test.

[willtebbutt]

LGTM. Thanks!