Add Dense/Sparse/Hybrid retrieval modes to the Qdrant vector store node,
using Qdrant's native BM25 sparse inference and RRF/DBSF fusion. Dense mode
remains fully backward-compatible when the new fields are left at defaults.

Retrieval modes:

• New "Retrieval Mode" field (Dense/Sparse/Hybrid, default Dense), plus
  Dense/Sparse Vector Name, Sparse Inference Model, and Fusion Method fields
• Sparse mode upserts directly via client.upsert(), skipping embeddings
  entirely
• Collection creation auto-configures named dense + sparse vectors in
  Sparse/Hybrid mode

Validation and error handling:

• Rejects unnamed dense vectors in Hybrid/named-Dense mode, and missing
  sparse vectors, with actionable error messages instead of silent
  fallback
• init() (query-only path) no longer auto-creates a missing collection in
  Sparse/Hybrid/named-Dense mode
• sparseInferenceModel throws on any value other than qdrant/bm25 instead
  of being silently ignored

Testing:

• 61 unit tests covering mode-specific schema, collection creation/
  validation, upsert wiring, query building, and response mapping
• Live-verified against a real Qdrant server: dense/sparse/hybrid queries,
  batch upsert, metadata filtering, Record Manager compatibility, and
  RRF/DBSF fusion

Documentation
FlowiseAI/FlowiseDocs#223

Closes #3007