Comparison during snapshot update should use `compare` function.

[jessemyers]

This line:

if encoded_expected_value is None or encoded_value != encoded_expected_value:

is using != rather than relying on the internal compare function, which means that this code cannot be adapted easily to types that do not support regular Python comparison, especially numpy arrays.

(Aside, it'd be especially nice if the Snapshot class allowed compare/encode/decode to be customized more easily; I'm currently experimenting with array comparison using numpy.allclose() by patching the internal _get_compare_encode_decode function.)

[olivier-plb]

Agree with the comment, at least compare function should be a parameter of assert_match and assert_match_dir, to allow better assert on binary snapshots (pandas.DataFrame, pyarrow.parquet, ...).
The default encode and decode functions are flexible enough, since we can encode to bytes prior the assert, but the compare function should be customisable to improve assertion error message like below:

E       AssertionError: value does not match the expected value in snapshot test_linear/linear.parquet
E         (run pytest with --snapshot-update to update snapshots)
E       assert b'PAR1\x15\x0...3\x00\x00PAR1' == b'PAR1\x15\x0...3\x00\x00PAR1'
E         
E         At index 1135002 diff: b'\xc0' != b'\xbc'

It should be straightforward to add an optional compare parameter to the assert functions. Otherwise, the _get_compare_encode_decode should be customisable. I believe an optional compare parameter should be more than enough.