fix: preserve Perl print literal boundaries in static analysis - #615

Referencing a Perl helper containing print "Use rm to remove a project\n"; currently produces HIGH AE1 findings even though the helper is readable and prints ordinary help text. The shell parser treats the closing Perl quote as an unmatched shell quote, records static_parse_limit, and propagates the partial result to every reference.

The customer reported this false-positive AE1 failure with SkillEvaluator 1.5.8 in NVCI run kpabani-ban-2026-09-23-001 (CI logs) and asked us to fix it. The report contains two HIGH AE1 findings for scripts/nvip_remove_project_iplite.pl, referenced at SKILL.md:52 and SKILL.md:351. A harmless Perl helper reproduces the reported failure signature; replay of the original customer helper is still needed to confirm its exact trigger.

Recognize a bounded form of standalone Perl literal print with LF or CRLF endings and preserve the ownership of its outer quotes. The entire payload remains subject to command scanning. Executable quote contexts, ambiguous package separators and sigil forms, interpolation, fragments, and genuine parser limits keep conservative coverage. Perl classification retains the existing hidden-instruction P2 and encoded-execution SC3 checks.

AE1 now carries a descriptive pattern, bounded target/analyzer/reason evidence, applicable limits, and targeted remediation. Individual references and incomplete-analysis gates remain intact. README and resource-bound documentation explain the diagnostics, supported Perl subset, and troubleshooting steps.

Validation:

• 162 Perl regressions and 516 broader static-pattern, runner, supply-chain, and finalizer checks passed.
• Independent review compared 330 TM1 combinations and 15 analyzers across 10 fixtures: no previously detected findings were lost after the applicability corrections.
• 19 native scanner-to-evaluator cases passed, covering harmless LF/CRLF help, executable removal, hidden instructions, encoded execution, ambiguous source, and real parser limits.
• Ruff lint/format and diff checks passed.

This is a bounded recognizer, not general Perl parsing support. The reproductions are synthetic and never execute their payloads. Downstream adoption requires a published release wheel containing the correction.