Fixed pivot in QR decomposition

[DurieuxPol]

The way pivot was handled didn't make much sense, so I reworked it a bit.
Now if you decompose a matrix m with pivot, you can reconstruct it with the permutation matrix:

m := PMMatrix rows: {
		     { 1. 2. 3 }.
		     { 4. 5. 6 }.
		     { 7. 8. 9 } }.
d := PMQRDecomposition of: m.
d decomposeWithPivot.
m closeTo: d q * d r * d permutationMatrixFromPivot inverse

The last line is equivalent to the equation M * P = Q * R.
I also removed the problematic test because it relied on the incorrect implementation of the pivot, and I added new tests instead.