Add class_path as a record type

[Simn]

Instead of a string list we now have a record with a path and a kind. Currently, the only kind is Directory because all I want for now is parity. Ultimately, I'd like to look into zip/archive class paths, and distinguish user/lib/std class paths explicitly.

[Simn]

To be honest, the diff here looks pretty awkward with all the path.Path.path business. I'll see if this can be encapsulated a bit better...

[Simn]

I ended up doing a bit more than I originally wanted, but all these should be good changes:

• class paths are now instances of ClassPath.class_path instead of raw strings, carrying some additional information and making this easily extensible. They all have their own readdir_cache, which was previously indexed by their path already anyway.
• Instead of matching against "std" and "_std" in various places, the compiler now tracks the scope of such class paths, which is Std and StdTarget respectively. This also led to the removal of com.std_path
• Similarly, class paths that come from haxelib have a Lib scope. This should make it possible to have something like -dce lib running on std and library, but not user code.
• The main file resolution function find_file now lives in ClassPaths.class_paths and has been greatly simplified. The file_lookup_cache also moved there.
• Resolved files are associated with the class path that was used to find them in ClassPaths.resolved_file. This distinction will allow consumers (that care) to adapt their behavior. For instance, the parser could read a file from a .zip archive instead of the disk.

Something else we could do now is associate modules with class paths. I'm not sure yet if this is useful for the compilation cache, but it's good to have the option.

@kLabz Would be good to test Shiro codebases with this, there's a chance that I missed some behavioral change.