[with patch] Wrap LinBox's Sparse Matrix Echelonizer over Finite Fields

[malb]

Apparently, LinBox? can compute echelon forms for sparse matrices over finite fields. And it seems to be faster than what we have now:

SAGE:

sage: A = random_matrix(GF(127),10000,10000,density=0.0002,sparse=True)
sage: time A.echelonize()
CPU times: user 99.64 s, sys: 0.22 s, total: 99.85 s

LinBox?:

matrix size :10000x10000
density = 0.0002
size before = 19981
Gaussian elimination (no reordering)...done (9.08057 s)
DONE
size after = 0 # Bug

I was told that SparseMatrixBase::NoReordering works but InPlaceLinearPivoting crashes.

Also, it claims to support GF(q) which is very very slow in SAGE right now.

[malb]

Actually, the above is not a bug but the size actually is zero after application of the Gaussian elimination. LinBox? clears rows it doesn't need anymore to save memory.

Also, InPlaceLinearPivoting does not crash if called correctly and is often faster than NoReordering.

[malb]

sparsegfp.patch requires  http://sage.math.washington.edu/home/malb/pkgs/linbox-20070915.spkg and exposes some of LinBox?'s functionality for sparse matrices over GF(p) to SAGE. More to come.