Separate fulltext engine extensions into "enrich" and "index" phases

Summary:
Ref T12819. Some of the extensions "enrich" the document (adding more fields or relationships), while others "index" it (insert it into some kind of index for later searching).

Currently, these are all muddled under a single "index" phase. However, the Ferret extension cares about fields and relationships which other extensions may add.

Split this into two phases: "enrich" adds fields and relationships so other extensions can read them later if they want. "Index" happens after the document is built and has all the fields and relationships.

The specific problem this solves is that comments may not have been added to the document when the Ferret extension runs. By moving them to the "enrich" phase, the Ferret engine will be able to see and index comments.

Test Plan: Ran bin/search index ..., grepped for indexFulltextDocument.

Reviewers: chad

Reviewed By: chad

Maniphest Tasks: T12819

Differential Revision: https://secure.phabricator.com/D18513