Clean up estimate_secondary ESS warnings in the prior choice guide and deduplicate per-chain fit warnings

[sbfnk-bot]

Description

Two small, related fixes for noisy diagnostics when fitting the auxiliary models.

Vignette — estimate_secondary() low tail-ESS. After the earlier warmup/sample bump, the prior choice guide render still showed low tail effective-sample-size warnings for estimate_secondary(). Reproducing the fit locally, bulk ESS already cleared at the fast settings (500 samples / 250 warmup) but tail ESS did not (min ESS ~384, just under the ~400 threshold). Giving the two estimate_secondary() examples a dedicated stan_secondary object (1000 samples / 500 warmup) clears it with comfortable margin (min ESS ~726, max R-hat 1.004). The change is scoped to the estimate_secondary() examples so the other models' rendered output is unaffected.

Logging — duplicated per-chain warnings. In the non-future fitting path, estimate_secondary() and estimate_truncation() fit all chains in one rstan::sampling() call, but the caught-warning handler in R/fit.R formatted each pooled warning with the whole chain vector, so sprintf recycled it and every message printed once per chain ((chain: 1), (chain: 2), ...). Collapsing the chain vector with toString() prints each pooled warning once as (chain: 1, 2). The future path (scalar chain id per call) is unchanged.

There is no dedicated issue for these; estimate_truncation()'s remaining ESS/divergence warnings are tracked separately.

Initial submission checklist

• My PR is based on a package issue and I have explicitly linked it. (no dedicated issue)
• I have tested my changes locally.
• I have added or updated unit tests where necessary. (behaviour is stochastic sampler logging; verified manually)
• I have updated the documentation if required.
• I have followed the established coding standards (checked with lintr).
• I have added a news item.

[sbfnk]

merging as docs only