Count-aware simulator and simulation smoke test

[m-murphy]

Parent

#45

What to build

Add count-aware simulation so users and tests can generate per-allele read count vectors from true within-host strain allele allocations, dropout probability, and noise rate—mirroring the count Poisson observation model in #48.

Expose R functions (extend or parallel existing simulate_observed_allele / simulate_observed_genotype) that return integer count vectors per locus rather than binary presence/absence. Parameters should align with count-model semantics (dropout on present alleles, Poisson noise on absent alleles, signal proportional to depth and support size).

Add a simulation smoke test: generate count data from known COI and error parameters → load with count aggregation → run short MCMC with observation_model = "counts" → completes without error. Qualitative recovery check optional (no strict stochastic tolerances).

Acceptance criteria

• R simulator generates per-allele read counts consistent with count observation model emissions
• Simulator documented in roxygen; exported or usable from test harness
• R testthat: simulated counts have expected shape and non-negative integers
• End-to-end smoke test: simulate → load (count mode) → run_mcmc(observation_model = "counts") passes with minimal burnin/samples
• Test skipped on CRAN if needed for runtime (consistent with test-mcmc.R)

Blocked by

• Count Poisson observation model — full MCMC inference path #48