Catch variance equal to zero when scaling

[dramanica]

This addresses #52
There are now checks both in big_scale(), as well as in big_randomSVD() and crossproductSelf() and tcroosprductSelf(). Unit tests have been added to all functions that were affected.

[dramanica]

@privefl I went for belt and braces. There is now a check in the scaling function (in case it is also used in another package), but also in the affected functions within bigstatsr. I guess this adds a redundant check if using big_scale() within a function that already has the checks (e.g. big_randomSVD, but the check is relatively light weight, so I don't think it is a problem (but you are free to disagree!)

[privefl]

Thank you for your work on this!

I've added a few changes:

• only test if it is 0 (instead of also negative) because I sometimes use negative scalings
• test for near 0 instead of exactly 0
• warn (instead of error) for big_scale() because I sometimes use this for computing the SD
• make sure big_randomSVD() also errors in parallel

[privefl]

Everything seems to work.
Tell me if you're okay with those changes.

[dramanica]

Fine with me!