Refactor parser code to remove tokenization code

[workmajj]

There's tokenization code leaking into the parser code (in particular, in parseLiteral()). We should clean this up so it's clearer what's happening with number literals, for example, where we'd expect any number (with underscores, periods, etc.) to come through as one token vs. several.

[workmajj]

When doing this, fix this bug too: 1 234 567 890 parses as &Number{1234567890, &NumberType{0}}, when really we should have an error (whitespace doesn't behave like underscores for our purposes).