Do not compute inverse for Blas Dixon

[jgdumas]

No need to compute the inverse mod p to lift in Dixon. The PLUQ factorization is sufficient to solve the system at each iteration.

[ClementPernet]

It may depend on the dimension/magnitude: although the overall arithmetic count is always in favor of the PLUQ variant, solving with a PLUQ at each iteration may be expensive: poorer cache behavious of the permutations and 2 trsv are less efficient than 1 fgemm (due to more modular reductions). Some benchmarking is required for a comparison.