Change ByteIdx in the lexer/parser to u32

[adamchristiansen]

Currently the ByteIdx into source files is a usize, which is unnecessarily large on 64-bit systems. This leads to a lot of wasted space in the AST when tracking source spans, since the upper 32 bits is never used (unless there is a source file >4 GB...). Switching to u32 will be more compact and still be able to represent any reasonable source file.