Optimize NSSS and CLI TTFX

[thorek1]

What changed

• Added automatic NSSS symbolic-mode selection for large redundant-variable workloads.
• Added reproducible command-line benchmarks for BasicRBC, SW07, FRBUS, and package-load timing.
• Added the generic first-order NSSS/IRF/moments precompile workload and a concrete IRF state-store type improvement.
• Documented timing methodology and measured results.

Why

SW07 and FRBUS spent a large share of first-call time in symbolic redundant-variable elimination during NSSS setup. The automatic mode preserves the symbolic path for small models and skips expensive symbolic solves for larger candidate workloads.

Validation

• SW07 automatic workflow: 46.88 s external wall time versus 53.39 s with explicit :single_equation in the same cached environment.
• BasicRBC benchmark passed with unchanged output checksums.
• FRBUS automatic workflow completed with the previously observed IRF checksum.
• Julia parsing and git diff --check passed.

SW07's documented undefined correlations are accepted as NaN; infinities and other non-finite outputs are rejected.