Remove docs matching hack and add more correct strategies for docs discovery.

[guineveresaenger]

This pull request includes changes to docs discovery that ensure we are matching entityDocs to the correct property field descriptions.

• adds number pattern to regex so we can match "S3"
• uses full parent path as docs path so we can properly map different "key" field descriptions to the correct parent field
• adds new regex to nested header discovery
• allows for dot-spaced keys in entityDocs for more precise lookup

Fixes #1789

Fixes #1876

Fixes #1875

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 60.13%. Comparing base (a38a786) to head (26fd8df).
Report is 18 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1882      +/-   ##
==========================================
- Coverage   60.99%   60.13%   -0.87%     
==========================================
  Files         303      327      +24     
  Lines       42457    44093    +1636     
==========================================
+ Hits        25898    26514     +616     
- Misses      15085    16090    +1005     
- Partials     1474     1489      +15     

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

[iwahbe]

Looks good to me.

We are waiting on an OK from the AWS folks to merge, right?

[guineveresaenger]

Yeah. I opened pulumi/pulumi-aws#3846 which only holds the schema diff when generated with d8ccf4c.

If we decide that that diff is too lossy for AWS, we have two options:

1. add a flag here - by contrast, Azure, GCP, and Cloudflare docs all look like they are very happy with these changes so we should definitely roll this out elsewhere.
2. implement Refine description matching when a description is unique #1886, see if that helps significantly for AWS and go through another round of checking with AWS squad.

Of course, we can also do 1 first and then implement 2 and remove the flag if deemed sufficiently better; if not, we could explore the option to flag resource-by-resource if top-used resources are negatively impacted.

[guineveresaenger]

The AWS folks say 🚢 since it this change removes a lot of incorrect docs, a net improvement for them. Thanks @t0yv0 for taking a look!