fix(math): preserve expm1 and log1p precision

[andrewtdiz]

Summary

• route direct Math.expm1 / Math.log1p codegen through the existing precision-preserving runtime helpers
• add parity coverage for tiny positive/negative values and 1e-9 sanity cases
• keep the naive exp(x) - 1 / log(1 + x) collapse contrast in the fixture

Fixes #2946.

Baseline

Pre-fix Perry lowered direct Math.expm1 / Math.log1p to exp(x) - 1 / log(1 + x), causing all six precision checks in the new fixture to print false.

Verification

• PATH=/home/github-runner/actions-runner/externals/node24/bin:$PATH node --experimental-strip-types test-parity/node-suite/globals/math-expm1-log1p-precision.ts
• target/release/perry test-parity/node-suite/globals/math-expm1-log1p-precision.ts -o /tmp/perry_math_precision_2946_post && /tmp/perry_math_precision_2946_post
• PATH=/home/github-runner/actions-runner/externals/node24/bin:$PATH RUSTC_WRAPPER= ./run_parity_tests.sh --suite node-suite --module globals --filter math-expm1-log1p-precision
  • report: test-parity/reports/parity_report_20260530_035423.json
• RUSTC_WRAPPER= cargo check -p perry-codegen -p perry-runtime
• RUSTC_WRAPPER= cargo build --release -p perry
• cargo fmt --all -- --check
• git diff --check origin/main...HEAD
• jq empty test-parity/known_failures.json test-parity/threshold.json
• rg -n "^(<<<<<<<|>>>>>>>)"
• ./scripts/check_file_size.sh