Optimize TypeScript code graph performance

[forhappy]

What changed

• Reuse a scoped preorder traversal and lexical indexes across TypeScript/JavaScript evidence passes.
• Avoid repeated resolver scans and skip JavaScript/framework resolution work when no relevant evidence exists.
• Fix AST fact-digest convergence and parallelize digest generation at the measured crossover.
• Use native atomic file copies for root artifact projection while preserving immutable snapshots.
• Add focused regression coverage and opt-in internal profiling boundaries.

Why

TypeScript/JavaScript code-graph builds spent substantial time repeating whole-tree traversals, linearly scanning evidence collections, and streaming immutable artifacts back to the root output. The fact-digest fixed-point loop also compared the wrong values, causing unnecessary work for nested scopes.

Impact

Published graph semantics and hashes remain unchanged for the qualified Axios and date-fns projects. Representative local runs improved, but the strict 2× performance target was not reached; this PR intentionally preserves exact graph output rather than changing the output contract.

Validation

• cargo fmt --all -- --check
• git diff --check
• sh scripts/check_product_boundary.sh
• CARGO_TARGET_DIR=/Volumes/Workspace/crabbuild-target/compass-42e2-performance cargo test -p compass-languages --test typescript_universal_candidate --locked — 113 passed
• Earlier focused resolver, framework, file-publication, digest, and Clippy checks passed.
• Axios and date-fns graph hashes matched their pre-change baselines exactly.