Consider making docs discovery schema aware

[guineveresaenger]

Our current strategy for discovering docs involves parsing upstream docs, detecting potential schema properties via patterns and headers/subheaders, gathering them in entityDocs and then at the end matching up the resource schema path to the docs path.

If we were to thread the resource schema through the docs generation process, we can be a lot more reliable in our docs discovery.

Issues such as pulumi/pulumi-libvirt#335 can be avoided because we could verify whether a found backticked item is a schema property vs a value.

#1960 - similarly, we'd not translate these into pulumiNames, because we'd know they're values.

Also, we could implement language-specific translation of properties inside field descriptions.

There's likely more. This is a fairly large refactor and should be planned carefully.