Fuzz DSLX parser

We should fuzz the DSLX parser to ferret out crashes and (to some extent) bad error messages. We'd be interested in random samples which are not valid DSLX programs. One way to do this is to take an set of existing DSLX programs and do random mutations such as: token removal, token insertion, token swapping and token mutation. In no case should the parser crash and it should always return an error message pointing to somewhere in the input. It would be interesting if the source span of the error message could be checked as well. I don't think we can guarantee that the span includes the mutated point of the program, but perhaps some related check could be performed.