Automate the calendar-year shifted-fit regression check in CI

[WestonVoglesonger]

Issue #49 deliberately separates the calendar-year behavior from automated two-fit regression coverage. The former gated comparison in #44 was removed because normal CI skipped it; the existing one-fit smoke test remains available for manual runs.

Goal

Add an automated CI check that fits two otherwise identical valid calendar-year workbooks with a fixed seed, shifting every observation and target year by the same constant (for example, +100 years). The internal model inputs should match; coverage estimates should agree within an appropriate numerical tolerance; reported birth years and the saved cohort origin should shift by that constant.

Work

• Measure runtime and setup cost in a clean CI runner, including imuGAP/Stan dependencies. A 100-iteration, one-chain calendar-year fit took about 0.6 seconds of run_fit() time on a local machine, but CI setup cost has not been measured.
• Build the smallest reliable two-fit regression test using two year ranges that both satisfy issue Support calendar years throughout imuRUN #49's 1000–9999 rule. Keep its sampling settings and seed fixed.
• Run it in a dedicated CI job or another explicitly scheduled automated CI workflow, with failures visible to maintainers. Document the trigger and measured runtime.
• Keep this heavier check separate from the normal fast unit suite and CRAN checks unless CI measurements support running it there.

This issue covers automated verification only. It does not change the calendar-year input contract in #49.