Make asset loading work cross-platform (Windows, *nix)

[augustl]

Note: this pull request depends on https://github.com/magnars/optimus/pull/66. The tests are running fine on travis/unix, but not on Windows, as it stands now.

I'll rebase the pull request against the JSR223 PR once it's ready. For now, this PR exists for quiet contemplation and discussion :)

• Changed optimus.paths to use java.nio.Path and make it work so that it does not rely on the paths on the OS level looking like URLs. On Windows, a \path\looks\like\this, so we parse it out, iterate the segments of the path, and generate URLs based on that. Removing dependency to "pathetic", the previously used path library.
• Changed optimus.assets.creation to support java.nio.Path based glob patterns, expressed as a vector. Also updated README to use these as the canonical path matcher, since regexps are tricky to make cross platform. Regexes are still supported, though.
• Parse java.class.path properly on Windows by using File/pathSeparator
• Changed digest code to explicitly use UTF-8 encoding. The tests assumes that the encoding is UTF-8, and this has no effect on deployments other than SHAs potentially changing without files changing if you happened to run your optimus code on e.g. Windows today.
• Changed optimus.export to close the IO streams it creates. It's good practice to close them, and Windows complains a bit more when they're not closed (the JVM is unable to delete the files associated with the streams)

[augustl]

I just thought of an alternative approach.

The file iteration can convert all paths on the class path to URL-like relative paths, before sending it through the asset loader. So, it would convert /home/foo/myapp/resources/js/test.js AND C:\Users\Foo\myapp\resources\js\test.js to /js/test.js - and then send it to the matching process.

This means we can leave out the change to a vector representing a glob, and keep using regexps as the default.

Thoughts?

[magnars]

Wow, well done on this! 👏

The second approach seems better to me. Normalizing paths as close to the problem as possible might also stop other bugs from popping up.

We could still support globbing, tho. The regexen are kinda weird.

[augustl]

I agree - didn't think about that until I went to bed right after making this pull request. Should have used mytomatoes.com... :)

I'll make two new pull requests: one for adding support for globs based on java.nio paths, and another for converting file system paths into URLs at the earliest stage possible, requiring the smallest amount of changes.