This release introduces flexible adapter resolution, new placeholder
adapters for OpenAI and Anthropic, and additional RAG strategies to
prepare for multi-provider support.

Adapter System Enhancements:

Add AdapterResolver module for dynamic adapter selection from context
maps or the PortfolioCore.Registry. This allows RAG strategies to
accept adapters at runtime rather than hardcoding specific
implementations.

Update Hybrid and SelfRAG strategies to use AdapterResolver, enabling
callers to pass custom embedder, vector store, or LLM adapters via
the context map.

New Adapters:

Add OpenAI embedder placeholder supporting text-embedding-3-small and
text-embedding-3-large models. Returns not_implemented error until
API integration is complete.

Add Anthropic LLM placeholder supporting claude-3-sonnet. Returns
not_implemented error until API integration is complete.

RAG Strategy Additions:

Add Agentic strategy placeholder that delegates to Hybrid retrieval
and tags results accordingly. Prepared for future agentic retrieval
logic.

Add GraphRAG strategy placeholder that delegates to Hybrid retrieval.
Prepared for graph expansion when graph adapters are configured.

Pgvector Improvements:

Make index creation idempotent by validating dimensions against
existing indexes rather than returning already_exists error.

Add normalize_metric and normalize_index_type helpers to accept both
string and atom inputs for metric and index_type configuration.

Application Configuration:

Add conditional child process loading via config flags. Set
start_repo, start_telemetry, or start_boltx to false to prevent
those processes from starting. Useful for testing and environments
where not all services are available.

Ingestion Pipeline:

Add enqueue/2 function for ad-hoc file indexing without starting a
producer. Enables programmatic indexing of individual files.

Dependencies:

Update portfolio_core dependency to version 0.1.1.