update docs for estimate_null_correlation

[stephens999]

can we document what this does by referencing @zouyuxin thesis when this is finalized?
presumably it is using the ad hoc em algorithm there?
does it also return a complete mash fit? ie do you have to run mash on this afterwards?
We should add it to the vignette on dealing with correlations.

I also note the phrasing "under the null" is confusing because it suggests we are assuming the null,
but actually it is the "residual" correlation (ie the correlation after removing effects).
I think the name estimate_null_correlation is fine, but let's clarify the description