Improve conversion between GAP and OSCAR finite fields

[fingolfin]

This is a change I started last April and never finished. I am resurrecting it now after a discussion with @mjrodgers and @ThomasBreuer

I wanted to refactor and possibly optimize the conversion between GAP and OSCAR finite field elements. While I got some modest improvements in one case, another got slightly slower. I also realized that there are some serious bottlenecks in Nemo that we probably should take care of. More on that below.

For now, some (old) timiings:

Before:

julia> FO = GF(3,4); FG = GAP.Globals.GF(3,4);

julia> f, finv = Oscar._iso_oscar_gap_field_finite_functions(FO, FG);

julia> e = one(FO) ; @btime f(e);
  1.875 μs (28 allocations: 832 bytes)

julia> e = one(FG) ; @btime finv(e);
  3.594 μs (38 allocations: 1.62 KiB)

After:

julia> FO = GF(3,4); FG = GAP.Globals.GF(3,4);

julia> f, finv = Oscar._iso_oscar_gap_field_finite_functions(FO, FG);

julia> e = one(FO) ; @btime f(e);
  1.429 μs (24 allocations: 624 bytes)

julia> e = one(FG) ; @btime finv(e);
  2.991 μs (39 allocations: 1.64 KiB)

One limitation is the conversion from a list of coefficients into an Nemo FFE (finite
field element). It looks like this:

function (a::FqField)(b::Vector{<:IntegerUnion})
  da = degree(a)
  db = length(b)
  da == db || error("Coercion impossible")
  return a(parent(defining_polynomial(a))(b))
end

So we take an integer vector, convert that into a polynomial, and feed that into C code which then copies the coefficients out of the polynomial, and then we discard the polynomial.

Perhaps @mjrodgers can file an issue at Nemo.jl about this, so we don't forget to improve this at some point.