Restore NB2 zero-mean padding in NegativeBinomial observation paths

[Copilot]

Zero-mean NegativeBinomial2 support in upstream Poisson handling did not resolve NB2 behavior in PyRenew’s observation paths. NegativeBinomialNoise and NegativeBinomialObservation still produced invalid (NaN) log-probabilities at exact zero means.

• Regression coverage for zero-mean NB2

  • Added focused tests for both APIs to assert finite log-probability when mean == 0 and obs == 0.
  • This closes the test gap that allowed the previous zero-padding removal to pass.

• Reintroduced epsilon mean padding in NB2 call sites

  • pyrenew/observation/noise.py (NegativeBinomialNoise.sample)
  • pyrenew/observation/negativebinomial.py (NegativeBinomialObservation.sample)
  • Both now apply epsilon padding before constructing dist.NegativeBinomial2.

• Documented numerical-stability rationale inline

  • Added short comments at both call sites explaining why padding is required.

padded_mean = jnp.asarray(mu) + jnp.finfo(float).eps
dist.NegativeBinomial2(mean=padded_mean, concentration=concentration)

Warning

Firewall rules blocked me from connecting to one or more addresses (expand for details)

I tried to connect to the following addresses, but was blocked by firewall rules:

• releases.astral.sh
  • Triggering command: /home/REDACTED/.local/bin/uv /home/REDACTED/.local/bin/uv sync (dns block)

If you need me to access, download, or install something from one of these locations, you can either:

• Configure Actions setup steps to set up my environment, which run before the firewall is enabled
• Add the appropriate URLs or hosts to the custom allowlist in this repository's Copilot coding agent settings (admins only)