Earley parser is extremely slow

[elliotchance]

The earley parser is designed for ambitious grammars by solving every possible path, but vsql only cares that there is one successful path so it just uses the first. However, this means that with an increasing number of tokens it gets exponentially slower.

We need to replace this with a LR(1) parser, but we'd need a generator like yacc for creating the tables and resolving the ambiguities.