improve testing of Filter class

[rabernat]

I implemented the filter class based on @iangrooms's original code. The filter class is responsible for picking the coefficients for the biharmonic and laplacian steps. It is tested in this file:

https://github.com/ocean-eddy-cpt/gcm-filters/blob/master/tests/test_filter.py

However, it would be best if someone who understands the procedure for defining the filter more deeply would help improve these tests. Right now I am just checking that a particular set of input parameters produces the same outputs I got in my own development environment, e.g.

gcm-filters/tests/test_filter.py

Lines 20 to 30 in 04ad835

	"filter_args, expected_filter_spec",
	[
	(
	dict(filter_scale=10.0, dx_min=1.0, n_steps=4),
	FilterSpec(
	n_lap_steps=4,
	s_l=[2.36715983, 8.36124821, 15.17931706, 19.7392088],
	n_bih_steps=0,
	s_b=[],
	),
	),

Is this sensible? Or can we think of more fundamental ways to test this part of the code.

[rabernat]

• Check that filter weights give good approx of target polynomial
• Check zero crossing

[rabernat]

As discussed in #20, we have a small gap in test coverage for different values of ndim here: https://codecov.io/gh/ocean-eddy-cpt/gcm-filters/pull/20/src/gcm_filters/filter.py?before=gcm_filters/filter.py#L70...82