feat: richer MCP tool surface — get_full_entity + progressive disclosure for agents

[rajnavakoti]

Description

Today the MCP server gives an agent essentially one mode: semantic retrieval that returns fused/partial context (score-weighted body fragments). That traps the agent in a single fuzzy mode and can't serve two common needs:

1. Aggregate/structured questions — e.g. "how many software systems are in this domain?" Semantic retrieval returns top-k similar chunks, not a complete count. (Covered by feat: query mode — let agents issue structured queries against the block ontology via MCP #74 query mode — cross-linked.)
2. Progressive disclosure / drill-down — the agent gets a partial snapshot, realizes it's relevant, and needs the complete entity. There's no way to say "give me the full entity."

The fix is an agent-native toolset the agent composes: discover fuzzy → drill to full entity → query for exact/aggregate. Fits the "primitives, not workflow" scope — hand the agent a toolbox, it decides.

The toolset

Tool	Purpose	Status
ask / retrieve	fuzzy discovery — find candidate entities	exists
get_full_entity(id)	return the complete entity (frontmatter + full body + relationships), not a fused snippet	this issue
list_entities(type) / count	enumerate/aggregate by type ("how many systems")	overlaps #74
get_related(id, relationship)	traverse the typed graph from an entity	this issue (optional)

Why (ties to architecture)

• Answers the exact/structured question class without an RDF/OWL back-end — the in-memory typed graph already knows entity types and relationships. See the "why not RDF/OWL" decision.
• Progressive disclosure means retrieval can stay lean (fused snapshots for discovery) while the agent pulls full detail only when it needs it — cheaper context, agent-controlled depth.

Acceptance Criteria

• MCP tool get_full_entity(id) returns the complete entity (frontmatter + full markdown body + resolved relationships)
• MCP tool list_entities(type?) returns entities (optionally filtered by type), enabling counts/enumeration
• (Optional) get_related(id, relationship?) returns entities connected to a given entity
• Tools are discoverable/described so an agent knows when to drill down vs. retrieve
• Tests for each tool + a discover→drill→query flow

Out of Scope

• Structured query language / query mode itself (feat: query mode — let agents issue structured queries against the block ontology via MCP #74)
• RDF/SPARQL back-end (deliberately not adopting — see why-not-RDF decision)