separate LU economy mode

[JeffBezanson]

Currently we can't infer a simple type for the result of lu since it might return either (LU, P) or (L, U, P).
I'm willing to give in on this in some cases, but this one doesn't seem necessary. Especially since lu(A, true) is not very easy to read. Yes, a keyword argument would be best but that still probably won't fix the type issue.
We need something like lu_compact or lu_econ.

Similar concerns might apply to other linear algebra functions. But it's much better if the various possible returns are prefixes of each other, i.e. (A,B), (A,B,C), (A,B,C,D) etc. so we always know the types of values that are requested.

[ViralBShah]

There are various forms of LU with various input combinations:

1. One output argument only, which does an in-place decomposition.
2. Two output arguments - L and U
3. Three output arguments - L, U, and P
4. A way to return P either as a vector or a matrix
5. In the case of sparse LU, there are 4 output args possible - L, U, P, and Q

It would be inelegant to have differently named functions for each case.

We certainly can ensure that the names are prefixes of each other, and that can be done in the case of existing lu also.

[JeffBezanson]

It's not as bad as you make it look. 4 is not necessary since the P vector can be turned into a matrix in an extra step. 2 and 3 can be the same function. 5 would be a separate method anyway since it is for sparse. So that really still leaves you with two functions for the dense case.