lm.fit() implemented for the 'more rows than columns' case

[Drew Schmidt]

lm.fit() is basically working now in parallel, provided you have more rows than columns (or the same number), even in the case where a rank deficient x is passed. 

This has passed initial testing, but more work needs to be done, not only in making it work for the remaining case (currently it will return an error because the numerical rank calculator doesn't do anything yet in this case), but also in testing.  This is currently experimental.  When it makes its way to the CRAN, I would consider it stable enough for mass use.

If anyone wants to take a look at it now, the code is up at my github:

https://github.com/wrathematics