[DNS] tp: simplify stdlib docs generation and make it more robust

The docs generator had some minor bugs with writing output to stderr
instead of propogating errors back to the caller. As part of trying to
fix this, I realised that the amount of state which is passed between
functions makes it very difficult to understand what is happening.

Moreover, there seemed to be quite a lot of duplication (i.e. repeated
parisng using regexes, merging annotations across multiple lines) o
subtle logic and unnecessary complexity in how we were finding the
boundaries of annotations.

To address this, rework the docs generator to instead be split into two
stages: a) extraction which is responsible for extracting the comments from
the SQL and "tokenizing" into description and annotations (including
merging sequential lines related to the same column) b) parsinng which
is repsonsble for actually verifying the semantics of the extracted
documentation and parsing into the JSON dictionary format expected by
the markdown generator

Bug: 283524256
Change-Id: Iccad2e64a4cb02411e39914d1b6cdbaa343a3611