[wptmanifest] Improve parser/serializer comment support

[jonathan-j-lee]

This change enables the metadata parser/serializer to preserve comments documenting test expectations. Currently, the parser ignores comments, which are lost when updating metadata from logs.

• The tokenizer now allows comments on their own lines.

• The tokenizer emits tokens for comments.

• The parser attaches comments to AST nodes. Comments are not AST nodes themselves, since operations on ASTs expect certain shapes.

• The serializer formats comment lines on a best-effort basis, like so:

  # <comment 1>
  # ...
  <node>  # <optional inline comment>
  

  The <node> can be a key, heading, condition, or list/value when used as the default after a list of conditions.

[jonathan-j-lee]

I'd like your thoughts on this change @jgraham. According to the blame, I think you're the expert on the test metadata format?

For some context, we're building a CLI frontend in Chromium that automatically updates expectations in bulk from wptreports generated by the CI system. Stripping out comments (as done currently) disallows documentation and creates noisy diffs.

[jgraham]

(also I note that wpt-update already exists for updating the metadata; are you using that or something else? We use it for Gecko and it seems to work OK, but the lack of roundtrip support for comments has been a problem, just never a high enough priority to actually fix)

[jonathan-j-lee]

(also I note that wpt-update already exists for updating the metadata; are you using that or something else? We use it for Gecko and it seems to work OK, but the lack of roundtrip support for comments has been a problem, just never a high enough priority to actually fix)

Something else that's Chromium-opinionated, although we delegate all the heavy lifting to the wpt backend.