[patch] [math] changes to QR decomposition

[Joni Salonen <jo...@gmail.com>]

Hi all, I've made some changes to the QR-decomposition code, namely:

- rewritten some comments for clarity
- extract the Householder transformation routine to a private method
- change the decomposition to "economy sized" in which Q is not always
square which saves memory in the m >= n case and seems to be how most
other libraries handle the decomposition
- changed some unit tests

Hopefully the extraction of the Householder transformation routine
makes subsequent modifications (e.g. adding solver routines) or
adaption of the code (e.g. for SVD which was recently discussed?)
easier.

Joni