fix(algorithms): use declared primary key instead of hardcoded node.id

[timothyEastridge]

Problem

GraphAlgorithms hardcoded RETURN node.id AS node_id (and {id: $x} MATCH filters), which raises "Cannot find property id for node" on any table whose primary key isn't literally id — i.e. essentially all real schemas (Store(store_key), Identity(identity_id), CustomerRecord(record_id), RetailOrder(order_id), …). This broke run_algorithm for the Bridgr agent on the seeded demo and any customer data.

Fix

• Add _pk(node_label) — resolves the declared PK via table_info (the Python binding reports the flag as "primary key", with a space); falls back to the first column, then id.
• Use node.<pk> in weakly_connected_components, pagerank, louvain, strongly_connected_components, k_core; use {<pk>: $x} in degree_centrality and shortest_path.
• k_core also returned the wrong column — core → k_degree AS core_number (verified against K_CORE_DECOMPOSITION output).

Verification

In-container against the seeded retail graph: Louvain over CustomerRecord/SPLINK_DB now returns non-null node_id (real record_id values); the agent's segmentation paints and traversal works.

Note

_build_adjacency (custom cross-label algos) still assumes id across heterogeneous PKs — left for a separate change; it isn't on the agent's run_algorithm path. bridgr-platform/bridgr-mono pins this commit; merging here keeps fresh clones + CI green.

🤖 Generated with Claude Code