fix(query): include unresolved bare-name targets in references_to

[nhuphuoc-bic]

Linked issue

No existing issue matched — I searched open and closed issues for references_to, unresolved targets, and workspace aliases and found none. Happy to file one via the issue form if you'd prefer that first.

What & why

references_to returns only the dependents whose REFERENCES edge carries a fully-qualified target (<file>::<Symbol>).

When a module imports a symbol through a specifier the parser cannot resolve — an npm/pnpm workspace alias such as @core/ports, or a tsconfig path mapping — the edge is still recorded, but its target is stored as a bare name. references_to never looks at those, so it drops the dependents silently. There is no confidence note on a non-empty result, so the short answer reads as a complete one.

That failure mode is worse than an empty result: the caller gets a small, plausible number and concludes nothing else depends on the symbol.

Measured on a pnpm workspace monorepo (TypeScript, apps/* + packages/* behind @core/* aliases):

symbol	references_to	actual dependents
an exported interface	6	12
another exported type	2	10

In both cases every missing dependent lived outside the defining package and imported through the alias. The edges were present in graph.db the whole time, under the bare target name.

callers_of already has exactly this fallback (the iter_edges_by_target_name pass that marks results target_resolution: "unresolved"). This PR mirrors it for references_to, with one added guard: bare-name edges are merged only when count_nodes_by_name(name) == 1, so a name shared by two symbols is never attributed to both. Edges carrying ambiguous_targets are skipped, as in the existing path.

After the change the two queries above return 11 and 9 — the remainder in each case is the barrel index.ts, which re-exports rather than references, so it legitimately produces no REFERENCES edge.

How it was tested

uv run pytest tests/ -q
# 2986 passed, 9 skipped, 2 xpassed in 58.47s

uv run ruff check code_review_graph/
# All checks passed!

uv run mypy code_review_graph/ --ignore-missing-imports --no-strict-optional
# Success: no issues found in 72 source files

The new test test_references_to_includes_alias_imported_dependents builds a fixture whose consumer imports through @core/ports, which the parser cannot resolve. Verified it is a real regression test: with the query.py change reverted it fails with assert 'usePort' in {}, and passes with the change applied.

Checklist

• Tests added for new functionality
• All tests pass: uv run pytest tests/ --tb=short -q
• ruff check clean
• mypy clean

[tirth8205]

Integrated on integration/token-efficiency-hardening as 48fc808. Unresolved unique-name references now remain visible, with a correction preserving target_resolution='unresolved' in minimal as well as standard output. Parser and query regressions cover both detail levels and ambiguous names; same-name ambiguity remains unassigned.