Implement lexer on mlstrings

The CakeML lexer operates on a `char list`. Therefore the first thing the CakeML compiler does after reading in its input string is `explode` the string into a list. This is wasteful: I think the lexer can easily be made to work on `mlstring`s.

This issue is to make the lexer operate on `mlstring`s and update the compiler accordingly to avoid the `explode`.

Extension thoughts:
- The `sexp`-parsing option would still need to `explode` the input because the `sexp` parser operates on `char list`s in a possibly more unavoidable way. (This issue does not require avoiding the `explode` for `sexp`s.)
- For large input, we might want to be even more sophisticated and lex/parse/typecheck incrementally, to catch errors before reading in all the source code. (Obviously this issue doesn't require doing that.)