feat: natural language search across the registry

[rajnavakoti]

Description

Replace or augment the current text-based search on the Discover page with semantic/natural language search. Users should be able to search with intent rather than exact keywords — e.g., "payment processing services", "things that talk to Kafka", "who owns customer data".

Acceptance Criteria

• Search input accepts natural language queries alongside keyword search
• Returns ranked results based on semantic relevance (not just string match)
• Searches across element names, descriptions, relationships, and field values
• Highlights why each result matched (which field/relationship matched the query)
• Falls back gracefully to keyword search if AI/embedding service is unavailable
• Works with the existing Discover page URL sync (query param preserved)

Out of Scope

• Vector database or external embedding storage (keep it simple, in-memory at build time)
• Real-time indexing (rebuild on deploy is fine for static site)