Handle sparse data robustly

[danielinteractive]

closes #3

Hi @friendly, thanks a lot for this nice package. I am proposing two small improvements in this PR:

1. Automatically omit strata with a single observation in CMHtest() because they do not contribute to the test statistics and currently cause the computations to return no results.
2. Use the generalized Moore-Penrose inverse from MASS in CMHtest() such that it can work when the variance
   matrix is singular.

I added also a bunch of tests for CMHtest. The test cases are courtesy of the https://github.com/PSIAIMS/CAMIS/ project, namely @statasaurus, @DrLynTaylor, @yannickvandendijck and @loganjohnson0. I have tried to appropriately call out their authorship in the test file.

Please let me know your comments, looking forward to your review :-)

Best, Daniel

[friendly]

Thanks for this, Daniel
The changes to CMHtest() you propose seem useful.
I haven't used testthat very much before and I don't use air, so I'm hoping this doesn't interfere with my setup.
I'll review this over the next few days.

[danielinteractive]

Thanks a lot @friendly !

No worries, the air config file won't interfere with anything else if you don't use air yourself.
The testthat usage should not be a problem and will just be part of the usual R CMD check.

[danielinteractive]

Hi @friendly , kind reminder to have a quick look at this if you get the chance, thanks!