Does the current charset detector actually even work?

[jfhbrook]

I have reasons to believe that the charset library that's intended to guess at charsets for content-types is actually just trying to parse them out of the existing content-type mimetype string, and so is effectively a noop.

I figure the way to fix this is:

1. Create a new charset abstraction that allows the same sort of registration as with mime types
2. Set up some sane defaults. This will require some research, and the answer might be "none".
3. Expose this functionality in the middlware/cli APIs and as a property on the ecstatic module