feat: extract GraphIndexMixin for shared merge/search/index

[Frankie-Xu]

Summary

• Extract merge_batch_data, build_index / build_node_index / build_edge_index, and search / search_nodes / search_edges into GraphIndexMixin beside GraphEditMixin.
• AutoGraph and AutoHypergraph both mix it in and drop their copies.
• Public search() still returns (nodes, edges). SearchHits stays in the CLI/MCP layer.
• Contract test: AutoGraph.search is AutoHypergraph.search so fix: forward source_ids/tags scope in AutoHypergraph.search #90-style source_ids copy-drift cannot recur.
• Rebased onto main @ 7a4db40. Commit author/committer is Frankie-Xu.

Maintainer accepted the local mixin direction on #115 (keep _node_memory / _edge_memory; do not push this into ontomem; GraphEditMixin stays the provenance/delete surface).

Closes #115

Out of scope

• Changing search() to return SearchHits
• RFC: migrate AutoModel/AutoList off pickle FAISS #116 pickle FAISS / RFC: chunk_rag as method-baseline eval vs graph_rag #117 eval harness
• Changing source_ids / tags semantics
• Moving the mixin into ontomem

Test plan

• uv run pytest tests/types tests/utils tests/cli tests/test_mcp_server.py (493 passed, 7 skipped)
• AutoGraph.search is AutoHypergraph.search
• Existing graph / hypergraph search-scope tests still pass

[yifanfeng97]

Merged — exactly what #115 called for: GraphIndexMixin beside GraphEditMixin, _node_memory/_edge_memory stable, net −168 lines, and a dedicated test file. Verified the full suite locally (611 passed). Nice follow-through from RFC to implementation.