Implement comment tracking in the new lexer

[mcy]

This change does two things:

1. It adds support for attacking comments to tokens in the token package, as well as a way to query the actual text of a comment token. Comment tokens which are token trees are used to represent tokens that are grouped together.

2. It adds support in the lexer for automatically attaching comments. I have included a test consisting of all of the funny cases listed on protobuf.com.

[jhump]

I don't think it's worth complicating the lexer this way.

In the current lexer, comment attribution is super dumb and simple. In order to implement trailing vs. leading comments in source code info, there is a later step that can decide to "donate" one token's leading detached comments to a prior token's trailing comments.

That also makes it way more predictable as far as what is a "trailing" comment in the AST, like for a formatter.

So in the current lexer, IIRC, a trailing comment is attributed only when the token is the last (non-skippable) token on the line and there is a comment between the token and the newline. For all other cases, the comments are associated with the following token.

Also, there was a fuzz test performance issue with having to calculate line numbers to do these checks. So I had to write #343 to work-around that. That causes the current lexer to track the current line number, just so it can annotate a token to keep track of when the previous token's end line matches the subsequent comment's start line (and to also make sure that the next token's start line is higher -- to make sure the two tokens are not on the same line).

I see your logic just scans the text for newlines, which is likely much faster than the logic I had that induced the perf issue (which used the O(log n) search to compute line numbers for the tokens and comments). But it still could be an issue for a pathological source file to have to do that scan instead of using trivial integer ops as each token is lexed to dead-reckon line numbers for this purpose.

[mcy]

As discussed elsewhere, W're going to punt this for now. I'm going to convert this to draft for future reference, but what we'll probably do is make comment attribution lazy.