Initial folder support

[mduvall]

This PR starts to address #7, a few questions:

• The input currently concatenates all the CSS files together and reports line numbers that aren't useful anymore past the first file). What's the ideal output format in this case? I see two ways of approaching this:
  • handle each CSS file individually (file name keys to current output blob)
  • handle them all together and add support in the output for mapping line number + file name (line: [1 (foo.css), 1 (bar.css)])
• Recursively look in directory for CSS files or top level of folder?

[Munter]

Isn't it better to just iterate over the files, keep the file names and line numbers seperate and thus usable in the output?

[mduvall]

@Munter Yeah that would be simpler, but how would we represent cross-file collisions in the output in that case (ie a collision in foo:1 and bar:2) and consistently map them assuming we need to coalesce the input?

[SlexAxton]

I think sourcemap support is the real win. We can take in a sourcemap if we need to, but we can also generate a sourcemap when we do the concatenation. Then no matter what we can take in a folder, a few random files, a stream of files, etc, and be able to map them to whatever we need without making the detection code too much more complicated.

The output would need some love, but my guess is that we could sort by file with most mentions (after sourcemap resolution), and only mention the first file or two that the collision occurs in (making sure to mention the file that the less used color is in). Generally you have a color that's used in lots of spots, and then a color that was used once in one spot, and collides with the color that should have been used. So as long as that color is easy to find, I think that's a win.

We can also have a --verbose mode for the full output of locations.