fix: resolve structural mutation ownership from scopes

[alxxjohn]

Summary

• replace guessed Go and C++ mutation ownership with declaration-, lexical-scope-, and bounded alias-based evidence
• keep unresolved symbols as non-baselinable diagnostics with per-language counts instead of promoting them to global/shared findings
• make structural fingerprints independent of diagnostic prose and evidence metadata while retaining pre-v2 baseline lookup compatibility
• preserve genuine receiver, argument, global, escaped, persistence, network, event, and shared-state findings
• add source-shaped regressions for crumb helpers, Overture resolvers, DuckDB hierarchy code, C++ builders, and DbRow::integer

Precision coverage

• Go imports, package globals, shadowing, nested closures, value/reference-backed fields, conversions, branch/loop state, and multi-hop aliases
• C++ locals, auto, constexpr, constructors, member initializers, references, pointers, moves, templates, lambdas, namespace-qualified owners, and compact assignments
• TypeScript nested-capture compatibility uses lexical visibility without restoring unknown-as-global guessing
• command/effect classification uses terminal-method and grammatical command evidence rather than repository allowlists

Verification

• go test ./... — 27/27 packages passed
• go vet ./... — passed
• golangci-lint v2.12.2 — 0 issues
• build and diff checks — passed
• independent task reviews and final scoped review — no remaining Critical or Important findings

crumb-app acceptance note

The read-only repaired scan reports 302 unsuppressed findings: 121 Go and 181 C++. It also reports 5,401 suppressed findings and unresolved diagnostics for both languages. Three genuine caller-owned Go findings were restored during final review.

The requested full-scan threshold of at most 19 cannot be evaluated as written against the supplied crumb-app revision/profile. A pre-repair control on the same checkout has 193 non-structural unsuppressed findings before structural findings are counted; 19 is the smell.message-chain count, not the full scan total. Baseline control reports 887 stale entries, while detector precision repairs necessarily change active/stale matching; the final read-only audit reports 5,267 active and 1,293 stale entries. No crumb files, suppressions, or baseline entries were modified.