Parser V2

[jprochazk]

Theres a few things about the current parser which make it not as nice as it could be. Specifically:

• It is not very resilient (only best-effort, using the sync helper on the parser to discard tokens in an attempt to return to a potentially good state)
• The indentation tracking machinery made it hard to implement semicolons
• The error handling mechanism is not great, because there's a distinction between "collect all errors" and "propagate this one error up"
• Spans are not associated with their source files/strings. Ideally spans would also be interned, as they currently take up quite a bit of space.

I think the issues are only going to become more exaggerated as we add features (match, do, with, postfix if/for, etc.), so before we start doing that, it's probably a good idea to do an overhaul of the parser while it's still this simple.

I think this post by matklad has excellent has ideas, and we should follow it when writing a new parser.