Require barry >= 0.2.2 (hash-collision fix) and make init_defm interruptible

[gvegayon]

Summary

Companion to USCbiostats/barry#24 and USCbiostats/barryr#8, which fix a catastrophic hash-collision bug in barry's FreqTable/vecHasher: support statistics (small-integer doubles) produced hashes with no entropy in their low bits, so unordered_map chained every key in a single bucket and support enumeration degraded from O(2^ncol) to O(4^ncol). In defm terms, init_defm() on a 16-outcome, order-1 panel model appeared to hang forever (~14 s per unique Markov state before, ~0.05 s after; a 20-state synthetic case went from >5 min to under a second). This PR:

1. Requires barry (>= 0.2.2) in LinkingTo so defm is compiled against the fixed headers (verified: R CMD INSTALL correctly refuses to build against barry 0.2.1).
2. Defines BARRY_USER_INTERRUPT as Rcpp::checkUserInterrupt() in every translation unit that includes barry, enabling barry's built-in interrupt checkpoint (every 1000 support iterations). Long init_defm() calls can now be stopped with Ctrl-C instead of force-killing R — this was the other half of the original report ("computer hangs and Ctrl-C does nothing").
3. Bumps version to 0.2.2.0 (tracking barry's versioning) with a NEWS entry.

Verification

• All 25 tinytest results pass against barry 0.2.2.
• Interrupt hook verified via setTimeLimit(elapsed = 3), which fires at the same R_CheckUserInterrupt checkpoint as Ctrl-C: a ~20 s init_defm() aborts at ~3 s with a clean R error (reached elapsed time limit), no crash or corruption on unwind. Plain Rscript ignores SIGINT entirely (verified with a pure-R loop baseline), so the time-limit route is the reliable non-interactive proxy.
• Timing unchanged on the support-enumeration benchmark (interrupt check adds no measurable overhead: 16-col/20-state init ~0.7 s, same as without the hook).
• Note: a model interrupted mid-init_defm() is left partially initialized and should be rebuilt with new_defm() before reuse — same caveat as any error thrown during init (documented in NEWS).

Merge order: needs barry ≥ 0.2.2 available (USCbiostats/barry#24 → USCbiostats/barryr#8 → this); CI will fail to satisfy LinkingTo until the barryr release is on CRAN/r-universe.

🤖 Generated with Claude Code

[github-actions]

Thank you for your contribution @gvegayon 🚀! The pkgdown site preview is ready for review 👉 Download here 👈!
_{(The artifact expires on 2026-10-13T19:10:49Z. You can re-generate it by re-running the workflow here.)}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 81.37%. Comparing base (57a7970) to head (e424fe4).

Additional details and impacted files

@@           Coverage Diff           @@
##           master       #8   +/-   ##
=======================================
  Coverage   81.37%   81.37%           
=======================================
  Files           8        8           
  Lines         451      451           
=======================================
  Hits          367      367           
  Misses         84       84           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.