feat(vectordb): Add database-agnostic vector search abstraction

[Shinu95J]

Add new vectordb package with Service interface for database-agnostic vector similarity search. This enables pharia- holmes to switch between vector databases (Qdrant, Weaviate, Pinecone) without code changes.

Changes:

• Add vectordb.Service interface with Search, Insert, Delete, and
  collection management methods
• Define DB-agnostic filter types: FilterSet, MatchCondition,
  MatchAnyCondition, NumericRangeCondition, TimeRangeCondition, etc.
• Add convenience constructors (NewMatch, NewFilterSet, Must/Should/MustNot)
• Implement custom JSON marshal/unmarshal for filter conditions
• Add type validation to prevent mixed types in MatchAny/MatchExcept

Qdrant adapter:

• Add qdrant.Adapter implementing vectordb.Service
• Add converter.go for vectordb → qdrant type conversion
• Expose Client() method on QdrantClient for adapter creation
• Remove duplicate types from qdrant/utils.go (now in vectordb)

Error handling:

• Changed Search() signature to return partial results: ([][]SearchResult, []error, error)
• results[i] contains results for request i (empty slice on failure)
• errs[i] contains per-request error (nil on success)
• Final error is for systemic failures (context cancelled, etc.)
• Uses sync.WaitGroup instead of errgroup to allow partial success
• Semaphore limits concurrent searches to 10

Package layout:

• vectordb/ - DB-agnostic interface and types
• qdrant/ - Qdrant adapter (implements vectordb.Service)