Index .mld page prose in sherlodoc

Remove Doc and Page kinds from the pure-documentation filter so that
page-level text content flows into register_doc's tokeniser. The
inner register_entry already skips type-search for untyped kinds and
skips full-name indexing for Doc, so no further changes are needed
for single-entry-per-page search.

Phase 1: one result per page, matched by any word in the page body.
Phase 2 (heading-level anchors) is a separate follow-up documented in
the plan.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>