fix bug in FswavedecnResult.__setitem__ and improve docstrings

[grlee77]

There was a bug in the dtype check in the setitem method of FswavedecnResult that prevents assigning modified coefficients. I think this probably warrants a 1.0.1 release to address.

The issue is that the user is supposed to be able to do something like:

dkey = (1, 2)
result[dkey] = modified_coeffs

A workaround in 1.0.0 is to do:

result.coeffs[result._get_coef_sl(dkey)] = modified_coeffs

This PR also improves testing of the FswaverecnResult and makes a few docstring improvements.

[grlee77]

It looks like numpy 1.9 did not support using assert_warns the way I had done in one of the new tests.

apparently use as a context manager requires numpy >= 1.11

[grlee77]

closes #424

[rgommers]

quick question: the dict -> list change is doc-only (was a doc error), and then sl -> tuple(sl) change is for clarity? I.e. neither is a backwards compatibility break?

[grlee77]

I don't think there is a backcompat issue. The tuple(sl) is just to avoid numpy 1.15+ FutureWarnings that weren't caught previously. That method should be private, so I don't think it is an issue to change the return type there.

[rgommers]

Okay thanks @grlee77 . LGTM, merged.

[rgommers]

I think this probably warrants a 1.0.1 release to address.

From a few days from now till the end of October I have almost no bandwidth. Feel free to just push some commits through if you want to get a bug fix release out.

[grlee77]

Thanks. I will probably create 1.0.x branch starting from the tag for 1.0.0 and then cherry pick the commits from this PR and #427. I will then plan to make a 1.0.1 release in the next week or so.