test(dead-code): cover find_dead_code end to end through the real parser→writer→query chain - #1598

@rrodriguesNutrium is attempting to deploy a commit to the shashankss1205's projects Team on Vercel.

A member of the Team first needs to authorize it.

CI is red on this PR, and I want to save a reviewer the investigation: the failures are pre-existing on main and not caused by this change.

The Build Test workflow is failing on main itself — run at 2026-08-13T17:58, before this branch existed. Same failures, same jobs.

Two distinct clusters, neither related to this diff:

1. tests/integration/cli/test_cli_commands.py — a CLI inventory mismatch over bundle inspect / bundle diff / bundle verify. The test's expected command set is out of step with the registered commands. Nothing to do with the writer.

2. tests/integration/test_parser_goldens.py (all languages) — these fail at

assert run_res.returncode == 0, f"Index/export failed for {project_name}..."

i.e. the index/export subprocess errors. That is an infrastructure failure, not a graph-content regression — the golden comparison never runs.

Worth being precise about, because this change does add a property (is_dependency) to contained nodes, so "all goldens failed" is exactly what a content regression would look like. It is not one. Locally, with this change applied, the goldens pass:

pytest "...test_language_golden[sample_project_go]" \
       "...test_language_golden[sample_project_java]" \
       "...test_language_golden[sample_project_c]" -q
3 passed in 75.73s

I ran the full 21-project golden suite green on a related branch too, so the baselines tolerate the added property.

One observation, offered as a suggestion rather than part of this PR: the golden test asserts returncode == 0 and bundle_path.exists(), but never that the exported bundle contains anything. When the index subprocess fails in a way that still exits 0 — FalkorDB Lite runs storage in a worker whose startup is timing-sensitive under CPU contention — the export writes an empty bundle and the diff then reports every node as missing. That failure mode cost me several hours: I concluded the golden suite was broadly broken, when in fact my machine was saturated by an unrelated index. On an idle machine all of them pass.

A guard between extraction and comparison — "indexing produced an empty graph, this is an indexing/database failure, not a parser regression" plus the subprocess stderr — would make that class of failure self-diagnosing. Happy to open it as a separate PR if useful.

The diff here remains 1 line of source plus a test; pytest tests/unit -q gives 12 failed, 1155 passed, all 12 verified pre-existing.

I know the CI is red, I have been merging some PRs and one of them broke this... Will have to check what's the issue before merging others.