Simplify and optimize FileMap's line/column mapping computation.

[eddyb]

Right now this is done while lexing:

rust/src/libsyntax/parse/lexer/mod.rs

Lines 423 to 435 in f861b6e

	if old_ch_is_newline {
	if self.save_new_lines_and_multibyte {
	self.filemap.next_line(self.pos);
	}
	self.col = CharPos(0);
	} else {
	self.col = self.col + CharPos(1);
	}
	if new_ch_len > 1 {
	if self.save_new_lines_and_multibyte {
	self.filemap.record_multibyte_char(self.pos, new_ch_len);
	}
	}

While that may seem like it avoids going twice over the file contents, lexing is slow, while newline and non-width-1 (which is also improperly computed by means of multi-byte characters, instead of the unicode-width library) character finding can be optimized one way or another.

Cleaning this up would get rid of hacks such as the save_new_lines_and_multibyte flag in the code linked above and having to reimplement this in anything else using CodeMap outside of libsyntax.

[wesleywiser]

I think @michaelwoerister made these changes in #50997. @eddyb do you think this can be closed or is there still more to do?

[michaelwoerister]

I think this should be pretty much fixed. Good catch, @wesleywiser!