To index notes, we exclude system notes and use `find_in_batches` to
load them in batches for submission to the ES bulk import API.
These queries often result in DB timeouts because the usage of
`ORDER BY id` results in the `notes_pkey` index being used.

This adds an optimized partial index, and removes the unused index
`index_notes_on_noteable_type` which is already covered for our
usage by the existing `index_notes_on_noteable_id_and_noteable_type`.

Newer versions of PostgreSQL (at least 11) are smarter about this and
use `index_notes_on_project_id_and_noteable_type` instead, so we might
be able to remove the partial index again in the future.