New fplll implementation

[Bill Allombert <Bill.Allombert@math.u-bordeaux.fr>]

Dear PARI developers,

We have added 3 variants of fplll to PARI, the fast, heuristic and dpe
variant. Internally, a wrapper selects a sequence of algorithms to
use.

Since a lot of PARI code rely on LLL, this can speed up a lot of
functions:

To give an example.

? qflll(matrix(500,500,i,j,random(100)),3);
Before:
  ***   last result computed in 29,201 ms.
Now:
  ***   last result computed in 4,889 ms.

Cheers,
Bill