Package housekeeping: R CMD check clean, silent tests, GHA CI

[smjenness]

Addresses three package-hygiene items before resuming the duration/dissolution work on #63.

1. R CMD check: (1 WARNING, 3 NOTES) → (0, 0, 0)

Issue	Fix
WARNING: undocumented build_epistats arg race.level	Add @param race.level with the allowed category values
NOTE: race.cat.num undefined global	Add to utils::globalVariables() in R/globals.R (dplyr::select(d, race.cat.num, ...) was flagged)
NOTE: hidden file inst/validation/snapshots/.gitkeep in build	.Rbuildignore the inst/validation/snapshots directory
NOTE: non-standard top-level CLAUDE.md	.Rbuildignore CLAUDE.md and .github/

Final check:

── R CMD check results ─────────────────────────────────────── ARTnet 2.9.0 ────
Duration: 31.3s
0 errors ✔ | 0 warnings ✔ | 0 notes ✔

2. Silent tabular testthat output

Root cause: tests/testthat/test-workflow-{standard,cea}.R were full ERGM estimation scripts — netest() + netdx() + print() + plot() — running as bare scripts (no test_that() wrapping). testthat executes every .R file in tests/testthat/ as test code, so they spewed hundreds of lines of SAN / MCMC progress per run.

Moved to tests/workflows/workflow-{standard,cea}.R with a tests/workflows/README.md explaining they're manual integration scripts. testthat::test_local() and the tests/testthat.R runner invoked by R CMD check only scan tests/testthat/, so the move cleanly excludes them without deleting functionality.

Also cleaned up a stale Rplots.pdf artifact from those scripts' plot() calls and added Rplots.pdf to .gitignore.

Before/after:

# Before: garbage for RStudio's test panel
⠏ |          0 | workflow-cea
Starting simulated annealing (SAN)
Iteration 1 of at most 3
...
[several hundred more lines per run]

# After: clean table
✔ |         77 | joint-dyad  [1.9s]
✔ |         78 | joint-model [1.3s]
✔ |         29 | joint-netstats [2.0s]
[ FAIL 0 | WARN 0 | SKIP 0 | PASS 184 ]     Duration: 5.3s

3. GitHub Actions CI for R CMD check

.github/workflows/R-CMD-check.yaml using the standard r-lib/actions template. Single-OS matrix (ubuntu-latest, R release) because ARTnet isn't CRAN-published — no need to burn minutes testing multi-OS portability. Runs on push to main, PRs targeting main, and manual dispatch.

ARTnetData in CI: installed, not skipped

ARTnetData IS the package's reason to exist, so tests that exercise it must run in CI. The default GITHUB_TOKEN can't see private repos in the same org, so the workflow uses a user-configured secret EPIMODEL_PAT.

One-time setup required in repo settings (commit 49b016e):

1. Create a fine-grained PAT with Contents: Read scoped to:
   • EpiModel/ARTnetData
   • EpiModel/EpiModelHIV-p
2. Add it to EpiModel/ARTnet → Settings → Secrets and variables → Actions → New repository secret, named EPIMODEL_PAT.

After that, CI installs both private packages, runs the full testthat suite including all joint-model / joint-dyad / joint-netstats tests, and runs R CMD check end-to-end. Suggests-gated skip_without_artnetdata() guards in tests stay in place as defensive no-ops (they skip gracefully in local dev if ARTnetData isn't installed, but never fire in CI once the secret is configured).

README gets a status badge linking to the workflow runs.

Test plan

• Local devtools::check(): 0 errors / 0 warnings / 0 notes
• Local devtools::test(): 184/184 pass, clean tabular output
• First CI run on this PR fails at dependency resolution (expected — EPIMODEL_PAT secret not yet configured)
• After EPIMODEL_PAT is added to repo secrets: re-run the workflow; should go green end-to-end with all ARTnetData-dependent tests actually executing

Bookkeeping only — no user-facing behavior changes. Byte-identical output for build_netparams() / build_netstats() is preserved and still verified by the inst/validation/ snapshot harness locally.