resources-graph: listings with exact and subsumed cache reads

[antejavor]

Part of PRD: resources-graph. Vocabulary: context-graph/resources-graph/CONTEXT.md. Prototype reference (throwaway): branch prototype/github-resource-graph-model.

What to build

Listings — the "go over 500 issues again" case.

• Parser: gh issue|pr list (state, label, author, assignee, milestone, search, limit), API listing paths, web ?q= → Listing Addresses with normalised filters; limit outside the key.
• Listing matcher (pure): exact match; subsumption only from a fully expanded Listing; free text exact-only.
• Store: shared (:Listing) with LISTS_FROM and HAS_MEMBER; members replaced on refetch; one Touch per listing.
• Sweep: page members at full depth, up to the asked limit; set fully_expanded.
• resource on a listing: paged index (number, title, state, labels, updated_at, comment count); subsumed answers say which listing and when.

Acceptance criteria

• All open memgraph/memgraph issues touched once → identical re-ask is a hit; --label bug is subsumed.
• A --limit 500 listing of a larger set is not fully expanded and never subsumes.
• Free-text searches only hit exactly.
• Matcher truth-table unit tests; e2e scenarios on real Memgraph.