perf(graph): bulk COPY FROM for nodes, routes, clients/producers (PR-P2)

[HumanBean17]

Scope

This PR implements PR-P2 from plans/active/PLAN-INIT-INCREMENT-PERF.md § PR-P2, converting the remaining graph write helpers to bulk COPY FROM:

1. _write_nodes_impl (shared workhorse) — stage Symbol rows then _bulk_copy(conn, "Symbol", _NODE_COLUMNS, rows)
2. _write_routes_and_exposes (shared) — per-table staging + _bulk_copy for Route/EXPOSES/Client/Producer/DECLARES_*/HTTP_CALLS/ASYNC_CALLS
3. _write_clients_producers_and_calls (incremental-only) — Client/Producer/edges bulk; MERGE (r:Route) retained + commented
4. _write_meta — left UNTOUCHED (on MERGE as planned)

Deletes _CREATE_SYMBOL, _MERGE_SYMBOL, _CREATE_ROUTE, _CREATE_EXPOSES, and 6 shared _CREATE_* constants (CLIENT/PRODUCER/DECLARES_*/HTTP_CALL/ASYNC_CALL).

Referential integrity handling

Applies the lesson from PR-P1's CI failure: bulk COPY FROM enforces referential integrity. This PR:

• Generalizes _existing_symbol_ids to _existing_node_ids (queries all node types: Symbol, Route, Client, Producer)
• Filters edge row endpoints (FROM/TO) against _existing_node_ids(conn) for EXPOSES, DECLARES_CLIENT, DECLARES_PRODUCER, HTTP_CALLS, ASYNC_CALLS
• Bulk-loads Route/Client/Producer NODES before edges that reference them (re-queries valid_ids after node bulk-load)
• Filters existing nodes in _write_nodes_impl to avoid duplicate primary key errors in incremental path

Column ordering fix

Enhanced _bulk_copy to select columns in exact schema order: tbl.select(columns). Without this, pyarrow's from_pylist creates columns in dict key order (alphabetical), which doesn't match LadybugDB's schema and causes type coercion errors.

Manual evidence

Full rebuild of tests/bank-chat-system with bulk writes:

rm -rf /tmp/p2-evidence && .venv/bin/python build_ast_graph.py \
  --source-root tests/bank-chat-system \
  --ladybug-path /tmp/p2-evidence/code_graph.lbug --verbose
.venv/bin/java-codebase-rag meta --source-root tests/bank-chat-system --index-dir /tmp/p2-evidence

Graph-write phase timing (from JCIRAG_PROGRESS lines):

• nodes written in 0.04s
• edges written in 0.15s
• routes/exposes written in 0.16s

Meta counts (from counts_json):

• Symbol nodes: 29 packages + 130 files + 140 types + 606 members = 905 total
• Routes: 29 (12 spring_mvc, 3 kafka, 14 builtin/layer_c_source)
• Clients: 8, Producers: 9
• Edges: EXTENDS 18, IMPLEMENTS 21, INJECTS 94, DECLARES 606, OVERRIDES 38, CALLS 678, EXPOSES 15, DECLARES_CLIENT 8, DECLARES_PRODUCER 9, HTTP_CALLS 8, ASYNC_CALLS 9

Tests

All tests pass:

• tests/test_incremental_graph.py — 30 tests (including 2 new: test_incremental_bulk_write_equivalent_to_full_rebuild, test_incremental_route_merge_dedup_preserved)
• tests/test_ast_graph_build.py — 28 tests
• tests/test_schema_consistency.py — 8 tests

Sentinel greps (all pass)

Must return zero:

grep -nE "_CREATE_(SYMBOL|ROUTE|EXPOSES|CLIENT|PRODUCER|DECLARES_CLIENT|DECLARES_PRODUCER|HTTP_CALL|ASYNC_CALL)\b" build_ast_graph.py
grep -nE "_MERGE_SYMBOL\b" build_ast_graph.py

Must be non-zero:

grep -n "MERGE (r:Route" build_ast_graph.py     # retained dedup (1 hit)
grep -n "MERGE (m:GraphMeta" build_ast_graph.py  # retained (1 hit)
grep -nEn "COPY .*FROM .rows" build_ast_graph.py # bulk present (23 _bulk_copy calls)

Files changed

• build_ast_graph.py — bulk conversions + column constants + _existing_node_ids helper
• tests/test_incremental_graph.py — 2 new tests in TestIncrementalOrchestrator