feat(search): tiered BM25 search (briefs/summaries/sources) + taxonomy accessors

[sebastianbraun25]

Problem

search_wiki (#233/#234) is a single BM25 index over concepts/, entities/, and
summaries/ combined. Two related issues surface as a wiki grows:

1. Concepts/entities shouldn't be keyword-searched at all. They are meant to be
   discovered by semantic browsing (an LLM scanning index.md's one-line briefs and
   picking the relevant slug by meaning), not by keyword frequency — but they currently
   share one BM25 corpus with summaries, so an unrelated but keyword-dense concept page
   can outrank the summary the caller actually needed.
2. One BM25 tier conflates two very different recall needs. A query for a generic
   term (e.g. "case") mostly needs a precision filter (which document's one-line brief
   is actually about that topic), while a query for a specific fact (e.g. an exact field
   name) needs recall into full document bodies — a single combined index tunes for
   neither well, and long PageIndex documents (sources/*.json) aren't searchable at all
   today, even though details like authorship/creation dates only ever live there, never
   in a generated summary.

Vorschlag

• Add list_taxonomy_items()/get_taxonomy_item() (openkb/agent/tools.py) for semantic
  browsing of persisted concepts//entities/ pages (kind-filterable), separate from any
  BM25 search.
• Add TieredWikiSearch (openkb/fulltext_index.py) with three independent BM25 tiers,
  each over a different granularity so a query only pays the recall/precision trade-off
  it actually needs:
  1. briefs — one-line description/legacy brief frontmatter per summaries/*.md.
  2. summaries — full body of summaries/*.md.
  3. sources — sources/*.md (whole file) + sources/*.json PageIndex docs, indexed
     per page (not per document), each hit carrying a Locator(kind="page", value=N)
     directly usable with get_page_content(doc_name, pages=str(N)).
• Add Locator (kind: "line"|"page") on SearchHit so a hit can point at an exact line
  (short docs) or page (PageIndex docs), not just "this file".

This issue covers the core engine only — no CLI/MCP/agent wiring (tracked separately as
follow-ups building on this one).

Kontext

• Builds on feat(agent): hybrid BM25 full-text search tool for query/chat agent #233/feat(agent): hybrid BM25 search_wiki tool for query/chat agent #234's WikiFullTextIndex/search_wiki (kept unchanged, byte-for-byte
  backward compatible, for existing callers) — TieredWikiSearch is additive, sharing the
  same BM25 math via an extracted _BM25Scorer.

---

_{This issue was drafted with the assistance of an AI assistant.}