[FEATURE] Add hybrid edismax + KNN reRank query mode (query.type=2)

[dkd-dobberkau]

Builds on #4065. Adds a hybrid (classical + KNN) sibling to the existing
query.type = 1 (pure vector) mode introduced as part of that feature.

Motivation

The existing plugin.tx_solr.search.query.type = 1 performs a pure KNN
vector search (q=*:* plus a vector frange filter), discarding classical
term-matching signals. For keyword-heavy queries this produces phantom
matches from the embedding cluster while exact-term hits get scored
identically to remote semantic neighbours.

In practice, on a TYPO3 demo site (solr-ddev-site with nomic-embed-text
embeddings), searching for WordPress in pure-vector mode returns 25 hits
(all unrelated TYPO3 pages sharing a "CMS" embedding cluster) while
searching for TYPO3 — the dominant term on the site — returns only 1.
Raising the minimum-similarity threshold does not help; it only shifts the
phantom count linearly.

What this PR adds

A third value query.type = 2 — hybrid re-rank mode:

• Classical edismax produces the candidate set (recall driver),
  preserving qf/pf/mm/highlighting/facets/spellcheck.
• Solr's {!rerank reRankQuery={!knn_text_to_vector ...}} orders
  the top-N candidates by bm25 + weight * cosine (precision refinement).
• Score-mix is tunable via two new TypoScript keys.

The hybrid branch is strictly additive: a new
isHybridVectorSearchEnabled() predicate sits inside the existing
else-arm of buildSearchQuery(). query.type = 0 (classical) and
query.type = 1 (pure vector) code paths are byte-identical to base.

Configuration

Two new TypoScript keys under plugin.tx_solr.search.vectorSearch.reRank:

Key	Type	Default	Effect
docs	Integer	200	Candidate pool size for re-ranking
weight	Float	2.0	Cosine multiplier: final = bm25 + weight * cosine

Existing keys (vectorSearch.minimumSimilarity, vectorSearch.topK)
stay relevant only for query.type = 1; the docs notes have been updated
to make this explicit.

Manual verification (TYPO3 14 / Solr 10)

Term	type=0	type=1	type=2
TYPO3	31	1	31
WordPress	0	25	0
language	6	varies	6

type=2 matches type=0 on recall (correct counts) while top-of-list
ordering is refined by the cosine signal. The phantom CMS-cluster matches
under type=1 disappear because BM25 = 0 documents are not candidates.

Test coverage

• 961 unit tests pass (8 new: getter defaults + configured values, hybrid
  build path including rq/rqq parameter shape, min() clamp branch
  with reRankDocs < topK, cross-relation with the existing vector flags)
• 84 integration tests (QueryBuilder|Search filter) pass, no regressions
• PHPStan level 6: 0 errors
• TYPO3 Coding Standards: clean

Known gaps / follow-ups

• The hardcoded model=llm f=vector strings in
  attachHybridVectorReRanking() mirror preparePureVectorSearch() — same
  hardcoding, kept for symmetry. A future PR could hoist both to a config
  getter.
• No integration test exercises hybrid against a live Solr — would require
  seeded vector fixture data, scoped out of this PR.
• The 2.0 default weight is a heuristic chosen for the demo site; large
  production indices may need re-tuning.
• Pre-existing doc inconsistency in TxSolrSearch.rst: section headings
  use query.vectorSearch.* while :TS Path: values correctly use
  plugin.tx_solr.search.vectorSearch.* (no .query. segment in the
  TypoScript path). The two new sections follow the existing heading style
  for consistency; cleanup is a separate concern.

Commits

1. [FEATURE] Add config getters for hybrid vector re-rank search
2. [FEATURE] Hybrid edismax + KNN reRank query path (query.type=2)
3. [BUGFIX] Clamp hybrid KNN topK to 1 and cover min() branch with a test
4. [TASK] Apply TYPO3 coding standards and fix PHPStan to hybrid vector path
5. [BUGFIX] Restore getRawSearchTerm() assertion with instanceof guard
6. [DOCS] Document hybrid query.type=2 and reRank settings