Do not treat unresolved same-name C# edges as confirmed callers, impact, or hotspot references

[Widthdom]

Summary

A qualified C# graph query can resolve the requested definition correctly and then attach unrelated calls that share only the leaf name. Those rows are reported as confirmed, non-heuristic graph evidence and also inflate hotspot counts.

This is a correctness problem: qualified navigation currently produces confident false positives.

Reproduction

Verified against the complete graph built from origin/main at 8d537125aae747c1008ad31b3d59364cb43652f0 with cdidx 1.42.0:

dotnet ./src/CodeIndex/bin/Debug/net8.0/cdidx.dll callers \
  ExclusiveFileLock.Open --exact-name --json --limit 10 \
  --db .cdidx/codeindex.db

dotnet ./src/CodeIndex/bin/Debug/net8.0/cdidx.dll impact \
  ExclusiveFileLock.Open --exact-name --json --limit 10 \
  --db .cdidx/codeindex.db

The definition resolves to src/CodeIndex/Cli/ExclusiveFileLock.cs::Open, but callers include, for example:

• PackageNormalizeCli.NormalizePackage, which calls ZipFile.Open;
• import/archive helpers whose Open calls target other APIs;
• DbContext paths that call SqliteConnection.Open.

impact reports these as confirmed_count: 10, hint_count: 0, and heuristic: false. inspect ExclusiveFileLock.Open exposes only the small set of actual references. The same unresolved leaf-name aggregation inflates symbol-specific hotspots counts for common names such as Open.

Expected behavior

• Resolved/confirmed graph rows must be tied to the requested symbol identity, not only callee_name == "Open".
• Unresolved same-name edges may be exposed as bounded hints, but must be labeled heuristic/unresolved and must not contribute to confirmed counts.
• Symbol-specific hotspot counts must use resolved identity. Any unresolved-name bucket must be separate and visibly lower confidence.
• Unqualified discovery queries may retain useful broad matching without weakening the contract of a qualified query.

Implementation guidance

Treat this as one PR because callers, impact, and hotspots consume the same edge-resolution contract.

• Prefer persisted or query-time target symbol IDs/container identity for uniquely resolved edges.
• Keep an explicit unresolved state; do not convert a leaf-name match to confirmed merely because the query root resolved.
• Share resolution-aware aggregation across graph traversal and hotspot ranking to prevent future drift.
• Define deterministic behavior for overloads, partial types, multiple definitions, external APIs, aliases, and incomplete extraction.
• Preserve bounded execution and expose counts for confirmed vs unresolved/hint evidence.
• Avoid dropping unresolved data globally: it remains valuable for broad search when honestly classified.

Likely areas include DbReader.GraphQueries.cs, DbReader.ImpactTraversal.cs, graph edge persistence/resolution, and hotspot aggregation.

Required tests

Add a compact C# fixture containing:

• two unrelated classes with the same method name;
• a qualified query for one class;
• an overload;
• a uniquely resolvable edge;
• an intentionally unresolved external edge.

Assert parity across references, callers, impact, and hotspots: only identity-matching rows/counts are confirmed. Also test incomplete-graph/cap metadata and unqualified fallback behavior. Add English and Japanese changelog fragments and update graph JSON contract docs if fields/semantics change.

Regression history

This is a direct residual/regression of the identity-resolution work in #4569 and the same-name overmatch across references/callers/impact/hotspots tracked in #3894. Earlier related work includes qualified graph queries in #2819 and hotspot same-name inflation in #151. All are closed; no open exact duplicate was found.

[Widthdom]

Work started on branch fix-issue5084.