Bind Node writes to one graph transaction

[eldonm]

Summary

• graph_transaction(db) opens the backend transaction, binds a GraphContext to that handle, and makes it the task-local default so Node.create / Node.connect participate.
• The old async_transaction_context opened a transaction but kept writing through the pool, so a rollback could not undo graph rows. It is now an alias for graph_transaction.
• Stores without begin/commit/rollback raise TransactionUnavailable.

Test plan

• tests/core/test_graph_transaction.py (commit, rollback, refuse)
• tests/db/test_postgres_integration.py::TestGraphTransaction against a live Postgres (skipped when DSN is down)

[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	0.027473	0.021299	-22.5%	OK
tests/benchmarks/test_deferred_save_benchmarks.py::test_bench_immediate_save_100	0.026193	0.020554	-21.5%	OK
tests/benchmarks/test_jsondb_benchmarks.py::test_bench_jsondb_batched_saves_500	0.496634	0.331186	-33.3%	IMPROVED (-33.3%)
tests/benchmarks/test_jsondb_benchmarks.py::test_bench_jsondb_count_empty_query	1.063421	0.701265	-34.1%	IMPROVED (-34.1%)
tests/benchmarks/test_jsondb_benchmarks.py::test_bench_jsondb_count_filtered	1.180206	0.872083	-26.1%	IMPROVED (-26.1%)
tests/benchmarks/test_jsondb_benchmarks.py::test_bench_jsondb_find_filtered	0.908534	0.654587	-28.0%	IMPROVED (-28.0%)
tests/benchmarks/test_jsondb_benchmarks.py::test_bench_jsondb_save_throughput	0.001719	0.001265	-26.4%	IMPROVED (-26.4%)
tests/benchmarks/test_sqlite_benchmarks.py::test_bench_sqlite_count_empty	0.345364	0.236069	-31.6%	IMPROVED (-31.6%)
tests/benchmarks/test_sqlite_benchmarks.py::test_bench_sqlite_count_fallback_via_regex	0.401450	0.285075	-29.0%	IMPROVED (-29.0%)
tests/benchmarks/test_sqlite_benchmarks.py::test_bench_sqlite_count_pushdown	0.376798	0.259902	-31.0%	IMPROVED (-31.0%)
tests/benchmarks/test_sqlite_benchmarks.py::test_bench_sqlite_find_fallback_via_regex	0.384202	0.286016	-25.6%	IMPROVED (-25.6%)
tests/benchmarks/test_sqlite_benchmarks.py::test_bench_sqlite_find_pushdown	0.344628	0.248999	-27.7%	IMPROVED (-27.7%)
tests/benchmarks/test_sqlite_benchmarks.py::test_bench_sqlite_sort_limit_pushdown	0.362602	0.291606	-19.6%	OK

[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	0.027473	0.026301	-4.3%	OK
tests/benchmarks/test_deferred_save_benchmarks.py::test_bench_immediate_save_100	0.026193	0.026433	+0.9%	OK
tests/benchmarks/test_jsondb_benchmarks.py::test_bench_jsondb_batched_saves_500	0.496634	0.431276	-13.2%	OK
tests/benchmarks/test_jsondb_benchmarks.py::test_bench_jsondb_count_empty_query	1.063421	0.939919	-11.6%	OK
tests/benchmarks/test_jsondb_benchmarks.py::test_bench_jsondb_count_filtered	1.180206	1.113885	-5.6%	OK
tests/benchmarks/test_jsondb_benchmarks.py::test_bench_jsondb_find_filtered	0.908534	0.818473	-9.9%	OK
tests/benchmarks/test_jsondb_benchmarks.py::test_bench_jsondb_save_throughput	0.001719	0.001545	-10.1%	OK
tests/benchmarks/test_sqlite_benchmarks.py::test_bench_sqlite_count_empty	0.345364	0.349407	+1.2%	OK
tests/benchmarks/test_sqlite_benchmarks.py::test_bench_sqlite_count_fallback_via_regex	0.401450	0.394677	-1.7%	OK
tests/benchmarks/test_sqlite_benchmarks.py::test_bench_sqlite_count_pushdown	0.376798	0.377409	+0.2%	OK
tests/benchmarks/test_sqlite_benchmarks.py::test_bench_sqlite_find_fallback_via_regex	0.384202	0.417122	+8.6%	OK
tests/benchmarks/test_sqlite_benchmarks.py::test_bench_sqlite_find_pushdown	0.344628	0.359279	+4.3%	OK
tests/benchmarks/test_sqlite_benchmarks.py::test_bench_sqlite_sort_limit_pushdown	0.362602	0.399144	+10.1%	OK