Fix false parser limits in Node assertion templates - #921

Node assertion diagnostics such as check.equal(actual, expected, `${row.key} is required`) were interpreted as shell command substitutions, producing a false static_parse_limit and failing strict completeness checks.

Add bounded recognition of direct diagnostic-template arguments on a verified node:assert or node:assert/strict import. Only the proven template delimiters change in the shell-completeness view; raw command checks, finding contents, and source offsets remain intact. Classify .mjs and .cjs consistently in ordinary files and archive members.

Unknown consumers, ambiguous syntax, aliases, shadowing, mutations, interpolation with calls or updates, fragments, and exceeded resource limits retain conservative behavior. TypeScript keeps its existing parser path because postfix and type expressions require different grammar. This covers a deliberately narrow JavaScript subset and assumes ordinary built-in assertion semantics; it does not add general JavaScript dataflow analysis.

Validation on the proposed commit:

• 2,538 relevant pytest cases passed, including 116 new regression cases.
• 11 isolated boundary, interruption, alias, and concurrency checks passed.
• Fresh base and HEAD wheels match every package/resource file in their source exports.
• 52 full-directory offline CLI scans cover 12 pinned sample skills plus the reported skill, on both revisions and in both source/wheel modes. All 26 source/wheel report pairs agree; all 12 comparison-skill reports, including their existing limitations, remain unchanged.
• The full reported skill was scanned on both revisions in source and installed-wheel modes: inspected coverage improves from 4/5 to 5/5 and strict CLI exit changes from 1 to 0. All four existing findings, confidence, risk score, and reference records are retained.

All verification above used isolated offline scans with --no-llm; scanned skill payloads were not executed. Target: main.