fix: connected-node fast path drops valid neighbors (0.0.11)

[eldonm]

Summary

Fixes two regressions in the 0.0.10 single-hop find_connected_nodes fast path that made .node(node=T) / .nodes(node=T, limit=n) silently drop valid neighbors. Found via an end-to-end failure of the embedded jvagent agent in integral (agent turns died with no_memory then session_resolution_error); root-caused to graph traversal, not agent code.

Bug 1 — limit-before-filter. limit was pushed into the DB scan before the Python node-type filter, so nodes(node=T, limit=1) returned nothing when a non-matching neighbor sorted first, even though matching neighbors existed. (In integral: an agent with 20 Memory neighbors → node(node="Memory") returned None → get_memory() failed.) Fix: push limit to the DB only when there is no node filter; otherwise apply after filtering.

Bug 2 — subtype resolution under an entity-name collision. The fast path deserialized every row with the base Node class, so when two Node subclasses persisted in one database share an entity name — e.g. an app User and an embedded-agent User — find_subclass_by_name(Node, "User") returned the first global match and _matches_node_filter's isinstance check dropped the valid neighbor. (In integral: conversation.node(direction="in", node=User) returned None → the session looked "foreign".) Fix: hydrate using the caller's requested concrete type as the resolution hint.

Why the suite missed it

The stock JsonDB exposes no find_connected_nodes, so every existing test fell back to the slow edge-scan path and never exercised the fast path. The new test installs a find_connected_nodes shim.

Test plan

• tests/core/test_connected_nodes_fast_path.py — 2 tests; both fail on 0.0.10, pass here.
• Existing traversal/graph suite: green (61 passed locally).
• Verified end-to-end: the integral embedded agent completes a full turn (reasoning + tool calls + answer) with this fix.

Bumps version to 0.0.11 + CHANGELOG.

🤖 Generated with Claude Code

[github-actions]

Benchmark comparison

Threshold: ±25% (informational, does not block merge)

benchmark	baseline (s)	current (s)	delta	status
tests/benchmarks/test_deferred_save_benchmarks.py::test_bench_deferred_save_batched_100	0.037416	0.045561	+21.8%	OK
tests/benchmarks/test_deferred_save_benchmarks.py::test_bench_immediate_save_100	0.037331	0.046178	+23.7%	OK
tests/benchmarks/test_jsondb_benchmarks.py::test_bench_jsondb_batched_saves_500	0.373731	0.610958	+63.5%	REGRESSION (+63.5%)
tests/benchmarks/test_jsondb_benchmarks.py::test_bench_jsondb_count_empty_query	0.799700	1.256767	+57.2%	REGRESSION (+57.2%)
tests/benchmarks/test_jsondb_benchmarks.py::test_bench_jsondb_count_filtered	0.921327	1.216108	+32.0%	REGRESSION (+32.0%)
tests/benchmarks/test_jsondb_benchmarks.py::test_bench_jsondb_find_filtered	0.705985	0.898894	+27.3%	REGRESSION (+27.3%)
tests/benchmarks/test_jsondb_benchmarks.py::test_bench_jsondb_save_throughput	0.001366	0.001868	+36.7%	REGRESSION (+36.7%)
tests/benchmarks/test_sqlite_benchmarks.py::test_bench_sqlite_count_empty	0.244740	0.312800	+27.8%	REGRESSION (+27.8%)
tests/benchmarks/test_sqlite_benchmarks.py::test_bench_sqlite_count_fallback_via_regex	0.330283	0.389930	+18.1%	OK
tests/benchmarks/test_sqlite_benchmarks.py::test_bench_sqlite_count_pushdown	0.273784	0.369238	+34.9%	REGRESSION (+34.9%)
tests/benchmarks/test_sqlite_benchmarks.py::test_bench_sqlite_find_fallback_via_regex	0.289234	0.385192	+33.2%	REGRESSION (+33.2%)
tests/benchmarks/test_sqlite_benchmarks.py::test_bench_sqlite_find_pushdown	0.242726	0.328375	+35.3%	REGRESSION (+35.3%)
tests/benchmarks/test_sqlite_benchmarks.py::test_bench_sqlite_sort_limit_pushdown	0.294411	0.359027	+21.9%	OK