Change the way we do Int64 conversion to/from GAP

[fingolfin]

Previously, any Int64 too big to fit into a GAP immediate integer was put into
a T_JULIA wrapper when passed to GAP, and unwrapped later. At the same time,
julia_to_gap relied completely (and incorrectly) on the C kernel functions
julia_gap and gap_julia to take care of all Int64 values.

The result was that passing such an integer as argument to a GAP function did
not work, not even if one explicitly used julia_to_gap.

To fix this we teach the C kernel function gap_julia to convert "large"
Int64 values to GAP big integers; this ensures that the many adapter functions
we install for Int64 arguments work.

This means that we violate the principle of roundtrip type fidelity (Int64 -> GAP large
int -> MPtr), but that is a very minor concern.

For really efficient conversions of e.g. Array{Int64,1} to GAP plists, we need
dedicated conversion functions which directly copy over bits. These can be
added transparently in the future.

Finally, we add some minor optimization to julia_to_gap for UInt64, Int128, UInt128

[codecov]

Codecov Report

Merging #357 into master will decrease coverage by 0.03%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master     #357      +/-   ##
==========================================
- Coverage   77.77%   77.74%   -0.04%     
==========================================
  Files          58       58              
  Lines        3995     3994       -1     
==========================================
- Hits         3107     3105       -2     
- Misses        888      889       +1     

[fingolfin]

@ThomasBreuer I have now completely rewritten this. Tests pass

[ThomasBreuer]

Looks good. Thanks.