Flx_pow with montgomery reduction

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

Hello PARI developers,

I have commited the change to Flx_pow to use Montgomery reduction.

Unfortunately I ran out of patience when trying to tune the
switch of algorithms.

The current tuning assume the Montgomery inversion take a negligible
amount of time (which is false if the exponent is small) and that p
is sufficiently small to use Flx_mulspec_mulii.

Also I have cleaned the Flx interface.

Cheers,
Bill.