Improve evalstr to show syntax warnings

[fingolfin]

Resolves #1058

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 74.97%. Comparing base (5cbad84) to head (379eb60).
Report is 1 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1059      +/-   ##
==========================================
- Coverage   75.04%   74.97%   -0.07%     
==========================================
  Files          55       55              
  Lines        4556     4560       +4     
==========================================
  Hits         3419     3419              
- Misses       1137     1141       +4     

Files with missing lines	Coverage Δ
src/GAP.jl	86.08% <100.00%> (+0.50%)	⬆️
src/ccalls.jl	99.20% <100.00%> (+<0.01%)	⬆️

... and 2 files with indirect coverage changes

[ThomasBreuer]

This is a good idea: Turn a real GAP error into a Julia error, and handle the GAP warning by just printing it.

As far as I understand, we have the following.
The syntax errors happen in the evalstr_ex call, and the buffer gets cleaned on the outer level, in evalstr.
Thus we will still keep the warning in the buffer if we call evalstr_ex directly. (Note that eval_str_ex is documented, and is mentioned in the docstring of evalstr.)

Logically, it would be better if evalstr_ex (or some helper function) would collect all GAP errors and warnings, clean the error buffer, and return also the collected error an warning information. Then the status of GAP would be clean after the evalstr_ex call. And evalstr would still be able to throw the Julia exception and to print the warning.

Is this reasonable?

[fingolfin]

I like the idea of adding the last error as a second argument to evalstr_ex. However it would be breaking change. @ThomasBreuer discussed this and agreed to merge this as-is for now, and we'll change evalstr_ex in a later PR for the next release.