Store line numbers in a more compact format

Currently there's a line number for every byte in the byte code, even though many bytes are going to correspond to the same line number. Since line numbers are only retrieved on error, retrieving a line number doesn't need to be particularly performant. Instead, they should maybe be represented as pairs of line number and bytecode range it corresponds to.

Start by locating where bytecode line numbers are stored and where error reporting retrieves them. Compare the current per-byte representation with a line-number and bytecode-range representation, then verify that error lookups still return the correct source line.

Written by the indexing model from the issue text.