perf(neo4j): fix CALLS write performance — backend detection, transaction scope, and MATCH optimisation - #1192

Summary

Three compounding issues caused slow CALLS edge writes on Neo4j, producing ~167ms/edge and multi-hour indexing times on large graphs.

Root causes fixed

1. Broken backend detection (writer.py)
The legacy getattr chain in write_function_call_groups returned the method object (always truthy) instead of calling it, silently using MERGE instead of CREATE on neo4j/nornic backends — negating PR #1161 entirely. Replaced with the canonical get_backend_type() call.

2. Single managed transaction across all CALLS batches (writer.py)
PR #1171's execute_write wrapped the entire _work closure — all 100k+ edges in one transaction. Neo4j held write locks and accumulated WAL entries until the final commit, producing 46-minute silent gaps with zero progress logs. Replaced with a plain session so each 1000-edge batch gets its own execute_write — per-batch auto-retry and clean rollback without the lock contention.

3. Inefficient MATCH on called nodes (writer.py, post_resolution.py)
The called MATCH used only name+path (2 of 3 composite index properties), causing Neo4j to scan all matching nodes and filter line_number in a WHERE clause: 26,680 DbHits per lookup vs 2 DbHits with the full composite index. Split each batch into:

• Fast path (called_line_number > 0): full 3-property composite index hit — 2 DbHits
• Slow path (called_line_number <= 0): ambiguous resolution, 2-property prefix scan (correct behaviour preserved)

Same fix applied to post_resolution.py where the caller MATCH had no label (full node scan across all nodes) and the called MATCH had the same 2-property issue.

4. SCIP cgcignore_path NameError (graph_builder.py)
cgcignore_path was referenced in _build_graph_from_scip but never passed as a parameter, causing SCIP indexing to always fall back to Tree-sitter with a NameError.

Benchmarks

Measured on a large Java Spring Boot service (~97k fn→fn CALLS edges), Neo4j backend:

Production baseline (v0.4.18): ~167ms/edge, 46-minute CALLS write gaps with zero progress logging. Fast/slow path split: 28% fast (known called_line_number, 2 DbHits), 71% slow (ambiguous resolution, index prefix scan). The fast path delivers 13,340× fewer DbHits per row — proportionally larger benefit on graphs with hundreds of thousands of nodes.

Test plan

• All existing unit tests pass (tests/unit/tools/test_graph_writer_transactions.py, tests/unit/core/)
• Pre-existing failure in test_database_kuzu_kotlin_metadata.py confirmed pre-existing on main, not introduced by this PR
• Full index run completed successfully with per-batch progress logging
• CI simulation (service re-indexed against a second large repo populated graph) completed
• SCIP fallback bug verified fixed — no longer errors with NameError