Parser clean-up

[travisbrown]

This is something I started last week—it's some improvements to the JavaCC parser that shave off a few hundred lines of code. The bulk of those come from rewriting the operator parsing to use "precedence climbing", to use the terminology of this article, instead of the "classic solution" (one non-terminal per precedence level).

This isn't really directly related to #1 or #2, but it makes editing the JavaCC grammar and digging through the JavaCC debugging output a little less painful. It should have no effect on the behavior of the parser.

As a side effect, it seems to speed up the parser a bit. For example, parsing the normalized prelude is around 6% faster in a benchmark I've included here. Before:

Benchmark                       Mode  Cnt   Score   Error  Units
ParsingBenchmark.parsePrelude  thrpt   20  15.305 ± 0.056  ops/s

After:

Benchmark                       Mode  Cnt   Score   Error  Units
ParsingBenchmark.parsePrelude  thrpt   20  16.235 ± 0.118  ops/s

The jar is also slightly smaller (104K instead of 108). Both of those improvements were non-goals, though, and I would have been happy to see either go the other way.