Change semantics of shebang in language configuration

[gkreitz]

Changes the semantics of shebang handling in language detection. When specifying a shebang, one must also specify a pattern (matching a subset or all files matched by files) in which one does shebang detection.

Shebang detection now only applies to language detection. Once a language has been selected, all files matching files in that language are included (shebang or not). In particular, this means that if a program has a foo.py with a python2-shebang and a main.py without a shebang, it will now be detected as python2, with two files, foo.py and main.py, where main.py is the mainfile. Earlier code would have detected this as python2_with_shebang, with a single file, foo.py (which would then also end up as the mainfile).

Note that our shebang detection is not perfect, and not quite in line with the wording of the spec (which I think requires a shebang in every .py file - Kattis/problem-package-format#591 ). In particular, if you have more .py files without a shebang than you do .py files with a python2-shebang, the language will be detected as python3 (both before and after this PR). Fixing that requires a larger rewrite, which I'm not sure is worth it (as incorrect language detection is extremely likely to be very noticable).

Fixes #457