feat(resolver): exploit confidence scores — resolution quality metrics

[zaebee]

Problem

confidence field exists on both Node and Edge models but is not exploited anywhere:

• resolved_edges returns everything — CONTAINS, IMPORTS, CALLS (resolved and unresolved) — with no quality signal
• There is no way to know "we resolved N out of M call edges" for a given ingest run
• cgis validate shows raw counts but not resolution rate

Proposed solution

1. Resolution rate in cgis validate — add a resolution_rate metric: resolved_calls / total_calls (exclude raw_call: targets)
2. Confidence-based filtering — add --min-confidence flag to trace/impact/structure commands to hide low-confidence edges from output
3. Ingest summary — print resolution rate table at end of cgis ingest (by edge type)

Acceptance criteria

• cgis validate outputs resolution rate (e.g. 78% of CALLS resolved)
• cgis trace --min-confidence 0.5 filters out unresolved raw_call edges
• Self-parsing test asserts resolution rate > threshold

Origin

Extracted from architecture feedback audit (PR #105, closed without merge).