fix(query): respect response_limit and report result_count for event/config patterns

[Alaa-Taieb]

Summary

In code_review_graph/tools/query.py, seven query-pattern branches (triggers_of, triggered_by, publishers_of, listeners_of, handlers_of, endpoints_for, consumers_of) bypassed the canonical add_result() helper. As a result they ignored the max_results / response_limit cap and never incremented total_results, so the returned result_count was always 0 while the response still contained N results — a contradiction that breaks pagination and any agent/UI relying on result_count / results_omitted.

This is Issue #674, Finding #4.

Fix

Route the seven branches through add_result() exactly like the sibling branches already do. Edges are still emitted unconditionally when the resolved node is None (preserving the prior contract — no edges disappear). consumers_of keeps its dedup and now only emits an edge for a newly-added result or a missing node, eliminating duplicate edges for already-seen consumers.

• No change to the add_result signature or any sibling branch — contained, low-risk diff.
• Does NOT touch Issue Bug: Codex installer generates Unix-only hooks on Windows #620 / PR fix(codex): emit commandWindows override for Windows hooks (fixes #620) #675.

Patterns fixed

Pattern	Edge kind	Direction
triggers_of	TRIGGERS	source → target
triggered_by	TRIGGERS	target → source
publishers_of	PUBLISHES	target → source
listeners_of	HANDLES	target → source
handlers_of	HANDLES	target → source
endpoints_for	HANDLES (Endpoint)	source → target
consumers_of	config consumer	config → source

Verification

• grep -n "results.append" code_review_graph/tools/query.py returns zero matches in the seven branches (only inside add_result and one unrelated branch at line 783).
• New regression test test_event_query_patterns_respect_max_results_and_report_count in tests/test_spring_events.py builds 5 publishers + 5 listeners against one event and asserts:
  • publishers_of(... max_results=3) returns ≤3 results, result_count == 5, results_omitted == 2.
  • listeners_of / handlers_of report result_count == 5.
• The new test fails on the old code (returns all 5, result_count == 0) and passes after the fix (negative proof done).
• ruff check clean; full tests/ -k query suite passes (22 passed).
• One pre-existing, unrelated test (test_event_resolver_does_not_cross_link_same_named_packages) fails on the Windows baseline regardless of this change; left untouched (it's a Windows path-separator issue in the test itself, not in the code under fix).

Manual test checklist (post-merge)

• query_graph("publishers_of", "event::X", max_results=3) returns ≤3 results but result_count == total publishers.
• query_graph("listeners_of" / handlers_of / triggers_of / triggered_by / endpoints_for / consumers_of) similarly report a correct (non-zero) result_count.
• No edges disappear from the edges array vs. pre-fix behavior (edges still emitted even when a node can't be resolved).

Closes #674.