Refactor example manipulation to use an AST

[iwahbe]

This PR converts the example processing section of *tfMarkdownParser.parse to use goldmark and an AST transformer.

[codecov]

Codecov Report

Attention: Patch coverage is 85.00000% with 42 lines in your changes missing coverage. Please review.

Project coverage is 57.14%. Comparing base (bdd39dc) to head (5a498cc).

Files with missing lines	Patch %	Lines
pkg/tfgen/docs.go	89.06%	13 Missing and 1 partial ⚠️
pkg/tfgen/parse/meta/meta.go	65.85%	10 Missing and 4 partials ⚠️
pkg/tfgen/parse/extension.go	88.65%	9 Missing and 2 partials ⚠️
pkg/tfgen/parse/section/section.go	78.57%	2 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #2312      +/-   ##
==========================================
+ Coverage   57.08%   57.14%   +0.06%     
==========================================
  Files         368      369       +1     
  Lines       50535    50676     +141     
==========================================
+ Hits        28848    28960     +112     
- Misses      20114    20136      +22     
- Partials     1573     1580       +7     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[iwahbe]

To merge, I think we need to add tests for splicing together external examples (I don't think that situation is under test at all right now). Otherwise I think this is ready to merge.

[guineveresaenger]

I'd love to get the typo fixed but this looks great!

[guineveresaenger]

nice find! I also like that the string renderer is refactored into the extension. 👍

[iwahbe]

I ran a downstream test run and reviewed the results. There are still wholes in our test coverage. I saw a bunch of positive changes, which makes me think this PR is worth merging. These include (but are not limited to):

Better parsing removes - - - from GCP (for every resource), inlining footnotes (necessary since we chop up the upstream document), and preventing the ## # Resource: ... pattern we see in many upstream providers.

---

Unfortunately, the PR also introduces some blocking regressions:

Previously, every title was Title Cased. We have accidentally dropped that behavior.
Links are rendered as absolute, so https://... is now rendered as [https://...](https://...). This is irrelevant to HTML but makes the markdown less readable.

Most concerning, some providers leaked their frontmatter into the upstream docs, for example newrelic and nomad:

• https://github.com/pulumi/pulumi-newrelic/pull/859/files: sdk/dotnet/DataPartitionRule.cs
• https://github.com/pulumi/pulumi-nomad/pull/407/files#diff-3954e5395e691069e93a8adf679180973d1475eb3092c2a25e738c72c1c332c7: sdk/dotnet/GetAclRoles.cs

[iwahbe]

Marking as a draft since this contains known regressions.