fix(go): field-chain calls resolve via validated type inference, never bare-name guessing

[colbymchenry]

Fixes #1276.

target.conn.Exec(...) where conn *sql.DB emitted a bare Exec ref (multi-hop receiver chains were dropped at extraction), and exact-match bound it to the only local Exec — the reporter's unrelated InternalStore interface.

What changed

• Extraction: Go 2-hop selector chains keep their receiver (target.conn.Exec).
• Resolution: a dedicated Go field-chain matcher resolves them via two validated inference hops — base's type from the enclosing scope (method receiver / typed param / local, the Local-variable receiver-type inference is C++-only — instance method calls through a local don't resolve in other languages #1108 machinery), then the field's declared type from the struct's own declaration lines (comment-stripped per line; chi's "the tree router" doc comment otherwise donates a phantom type). resolveMethodOnType validates the result.
• Policy: this path is exclusive for chained Go receivers. External field types (sql.DB, http.Handler) produce no edge — exactly the issue's ask — and package-qualified types are only followed when the package is in-module, so a stdlib interface can never bind a same-named local decoy. No fall-through to bare-name strategies (chained refs were never emitted before, so no existing recall lived there).

Verification

• Tests: the issue's exact fixture (no InternalStore edges; Repo.Save → Store.Put in-project chain resolves), plus the unexported-type (tree *node) recall case, the doc-comment guard, and the http.Handler-decoy guard.
• Real-repo before/after on go-chi/chi: node count stable (1,181); +8 correct, validated field-chain edges (mx.tree.FindRoute/InsertRoute/routes, r.MatchOne.Match — all previously unresolvable or wrong); removed edges are the prior bare-name guesses on external receivers (r.Header.Get → Mux.Get and kin).
• Full suite: 145 files, 2439 passed.

🤖 Generated with Claude Code