Weibull delay support + distspec alignment (0.2.0)

[sbfnk-bot]

Prepares the 0.2.0 release, folding in Weibull delay support (per the discussion) and aligning with the soon-to-release distspec.

Weibull delays

fit_cfr() and simulate_linelist() now accept a Weibull() onset-to-death (and recovery) delay, alongside LogNormal() and Gamma(). The Stan cure model is family-agnostic (it reads brms's parameterisation and gets the delay CDF from primarycensored's dist_id), so no new Stan code was needed — the work is R-side:

• .assert_delay_family(): allow weibull.
• .delay_family_prior(): map distspec's (shape, scale) to brms's (mu = mean, shape) via .weibull_mu_prior() (mean = scale·Γ(1+1/shape), delta-method uncertainty) — the analogue of the existing gamma reparameterisation. Refactored the family branch to a switch().
• sample_delay(): rweibull. .delay_moments(): Weibull mean/sd.

Verified end-to-end: a Weibull fit recovers the CFR (0.398, 95% CrI covers 0.4) and delay mean (11.0, covers the true 11.7), all rhat < 1.05.

distspec alignment

distspec now exports natural_params(), which cfrnow reimplemented privately as delay_native_order(). Adopted natural_params(delay) (dispatch on the object) and removed R/utils.R. cfrnow's full suite already passes against the new distspec; this just tracks the public API.

Release

Version → 0.2.0; NEWS rolled into a 0.2.0 section (Weibull, distspec, stratified vignette, plus the existing pp_check/ascertainment/grouped-CFR entries).

Post-merge (separate steps): tag 0.2.0, add cfrnow to the epiforecasts r-universe registry, then update the README install instructions.