Stop including end comments in collections if next content is not sufficiently indented

[eemeli]

Fixes #41

Effectively, when encountering a comment on a new line while parsing a collection, the parser now peeks ahead to see if the next non-comment content is indented at least to the level of this collection, or if there's a subsequent comment that's sufficiently indented. This means that e.g. parsing and re-stringifying this document:

#c0
- #c1
  k1: v1
  #c2
  k2: v2 #c3
#c4
  k3: v3
#c5

Results in this representation:

#c0
#c1
- k1: v1
  #c2
  k2: v2 #c3
  #c4
  k3: v3
#c5

Previously, the #c5 comment would have gained a spurious level of indentation due to being incorrectly attached to the k* map rather than the root sequence.

@ikatyang Did you have some fixes in yaml-unist-parser for this following prettier/prettier#4917 that would need to take into account these changes?

[ikatyang]

Thanks for the notice. Basically, yaml-unist-parser re-attaches every comment (with a few exceptions) so the result should be the same as long as all the comments are collected correctly. I just tried this branch in yaml-unist-parser, all the tests are passed.

I think this kind of comment attaching can be hard for a parser since it needs to be fixed one-by-one, so I didn't raise this sort of issues here. Instead, I apply a postprocessing for comment attaching in yaml-unist-parser, which I just need to define general rules (something like "when A comment is in front of B node, attach A to B") and it'll work for all the cases.