Search: preserve operation id (e.g. _bulk) in OpenAPI search_title

[Mpdreamz]

Summary

For OpenAPI-exported docs (e.g. the Elasticsearch Bulk API, operation id _bulk), search_title currently reads "Bulk index or delete documents - Elasticsearch API - doc operation" — the operation id never appears, and its leading underscore would be lost even if it did. Users search Elasticsearch docs by REST endpoint name a lot (_bulk, _search, _cat, dotted ids like ml.start_datafeed), so this is a real, missing search token.

Root cause: OpenAPI docs are built in two stages, and the second silently overwrites the first's search_title. OpenApiDocumentExporter sets Title == SearchTitle, but ElasticsearchMarkdownExporter.CommonEnrichments then unconditionally rebuilds SearchTitle via CreateSearchTitle(), which derives extra tokens from the URL by splitting on /, , -, ., and _. For the Bulk API (URL segment operation-_bulk), that split turns _bulk into a bare bulk, which then gets deduplicated against the title's existing Bulk token and dropped entirely.

Fix: let OpenAPI docs own their SearchTitle.

• CommonEnrichments now only rebuilds SearchTitle via CreateSearchTitle() for non-API content (doc.ContentType != "api") — markdown docs are unaffected.
• OpenApiDocumentExporter builds a dedicated search title appending the raw operation id verbatim (underscore/casing intact): "Bulk index or delete documents - Elasticsearch API - _bulk". Also drops a stray trailing space in Title that caused the reported double space.

website-search-data check: no mapping/analyzer change needed. search_title is analyzed by synonyms_fixed_analyzer/synonyms_analyzer, both using group_tokenizer, whose delimiter set doesn't include _ or . — so _bulk (and dotted ids like ml.start_datafeed) index and search as a single token, both index- and search-time, including on the completion (search_as_you_type) sub-field. Flagged separately, not blocking: title.starts_with's EdgeNGram tokenizer token_chars omits "punctuation" and would strip a leading underscore if operation ids are ever added to Title too.

Test plan

• dotnet build for Elastic.Markdown and Elastic.ApiExplorer succeeds
• New OpenApiSearchTitleTests (Elastic.Markdown.Tests) — asserts CommonEnrichments preserves the API exporter's search_title and still rebuilds it for markdown docs
• New OpenApiOperationIdSearchTitleTests (Elastic.ApiExplorer.Tests) — exercises OpenApiDocumentExporter.ConvertToDocuments against an in-memory Bulk API spec (no network), asserts search_title == "Bulk index or delete documents - Elasticsearch API - _bulk"
• Full Elastic.Markdown.Tests and Elastic.ApiExplorer.Tests suites run; pre-existing failures are unrelated ScopedFileSystem/git-checkout tests failing in this worktree environment
• dotnet format / lint clean (pre-push hook passed)
• Coordinate a reindex with the website-search-data owner once merged, alongside the companion relevance-tweaks PRs (Search: sync agg/aggs synonym with website-search-data #3606, Search: guard navigation_* rank features against penalty default #3607, Search: bump contract to 0.12.0 and populate content_tier #3608)