Speed up cperm

[fingolfin]

Also adjust the docstring. Point out that omitting the parent group is not advisable, and change uses of cperm in tests to provide a parent.

Before:

julia> g = symmetric_group(15);

julia> @btime cperm($g, 1:4,5:6,7:15);
  5.007 μs (69 allocations: 3.05 KiB)

After:

julia> g = symmetric_group(15);

julia> @btime cperm($g, 1:4,5:6,7:15);
  1.238 μs (6 allocations: 416 bytes)

There is actually some more space for improvement by adding some low level primitives for accessing and creating GAP permutations, but that's for a future PR. In the meantime this already improves things and I don't want to rebase this once again.

This was created as part of work with @Pascalonyy on Origamis from 1 year ago, for AG Weitze-Schmithüsen in Saarbrücken. Sadly that never went anywhere as far as I know, but I can at least salvage some of the bits I provided to make things much faster there.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 84.37%. Comparing base (6c0ad05) to head (733420e).
Report is 197 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #4361   +/-   ##
=======================================
  Coverage   84.37%   84.37%           
=======================================
  Files         656      656           
  Lines       87190    87202   +12     
=======================================
+ Hits        73563    73575   +12     
  Misses      13627    13627           

Files with missing lines	Coverage Δ
src/Groups/perm.jl	98.84% <100.00%> (+0.09%)	⬆️

... and 1 file with indirect coverage changes

[lgoettgens]

Triage likes the idea that cperm delegates to a less efficient function in the case that it detects overlapping cycles.

[ThomasBreuer]

Currently the documentation of @perm also states that overlapping cycles are supported, and mentions cperm.

[fingolfin]

I have now updated this to restore support for non-disjoint cycles.

[ThomasBreuer]

Looks good, thanks.
(A local variable gens is used in the tests, but this can be improved in another pull request.)