__array_function__ test coverage; numpy 1.17; result_type()

[crusaderky]

• Support for numpy 1.17 (it's already what's used in CI today)
• Implement sparse.result_type + unit tests
• Refactor environment variables and other switches into sparse.settings
• Add unit tests for mixed numpy+sparse inputs to __array_function__, with different permutations

[codecov]

Codecov Report

Merging #272 into master will increase coverage by 0.05%.
The diff coverage is 92%.

@@            Coverage Diff             @@
##           master     #272      +/-   ##
==========================================
+ Coverage   94.73%   94.78%   +0.05%     
==========================================
  Files          14       15       +1     
  Lines        1955     1976      +21     
==========================================
+ Hits         1852     1873      +21     
  Misses        103      103

[hameerabbasi]

I added a nice test for NEP-18 here: dask/dask#4675. Would you care to copy that one, as it tests the protocol?

[crusaderky]

@hameerabbasi done

[hameerabbasi]

Can you change the implementation of result_type to the following:

return np.result_type(*(x.dtype if isinstance(x, SparseArray) else x for x in args)

And move it to common.py?

[crusaderky]

All done, see if you like it