Arbcall refactor

[saschatimme]

This refactors the arbcall logic to fix #55 and to hopefully avoid things like this happening in the future.
This also adds some more features:

• Defines constants for the type unions we consider equivalent, e.g., const ArbLike = Union{Arb,ArbRef,Ptr{arb_struct}}.
• Overloads the show method for these constants to get more readable output:

# before
  Closest candidates are:
    add!(::Union{ArbVector, Arblib.arb_vec_struct}, ::Union{ArbVector, Arblib.arb_vec_struct}, ::Union{ArbVector, Arblib.arb_vec_struct}; n, prec) at /Users/sascha/code/Arblib.jl/src/arbcall.jl:275

# now
  Closest candidates are:
    add!(::ArbVectorLike, ::ArbVectorLike, ::ArbVectorLike; n, prec) at /Users/sascha/code/Arblib.jl/src/arbcall.jl:275

• The detection of length arguments is better now. If the argument is an Int and the previous argument ArbVectorLike or AcbVectorLike then the name of this argument is moved to be a kwarg, but only if the name is n or starts with len.
• I also enabled in a separate commit the possibility to automatically also generate the true arb signature, i.e., without moving any positional argument to a keyword argument.

This also probably needs some more tests, but this not a task for 10pm :D

[kalmarek]

I'm happy with the changes, but @Joel-Dahne have a look at this as well

[saschatimme]

@Joel-Dahne @kalmarek and myself are both happy with the changes and think that it's a good idea to always also generated the original signature as an escape hatch if our replacement logic is non-sensical. What do you think?

[Joel-Dahne]

One issue with the automatic length is that you can still pass in two vectors with different lengths and crash Julia. So all of these methods should be considered unsafe. If we ever add exported versions of them we need to add proper bounds checks when wrapping them.

[Joel-Dahne]

Could we factor this out as a separate method? Mainly to make it easier to test independently since this is quite delicate.

[saschatimme]

I factored it out and added some tests

[Joel-Dahne]

Minor thing, but can we either keep the Ref versions just under the default version or have all of them in the end, to make it easier to see directly what is going on.

[saschatimme]

Done

[Joel-Dahne]

I did a very quick review of the code. In general I think it looks good, less code now. I think generating the original signature as well is reasonable! If we merge this then I can update #32 before merging that.