feat: Implement Barlow-Beeston lite for staterror modifier

[kratsg]

Implements BB-lite for the staterror modifier in pyhs3's HistFactory
distribution. BB-lite uses shared gamma parameters across samples with
a channel-level constraint built from combined MC statistical uncertainties.

Key changes:

• Add barlow_beeston_method field to HistFactoryDistChannel (defaults to lite)
• Make StatErrorModifier.data optional (None for lite mode)
• Allow constraint: Poisson in addition to Gauss for staterror
• Implement _make_barlow_beeston_lite_constraint() supporting both Poisson and Gaussian
• Modify extended_likelihood() to skip per-sample constraints in lite mode
• Add comprehensive test suite (12 tests) for BB-lite functionality

BB-lite vs BB-full:

• BB-lite (default): Shared gamma params, channel-level constraint from combined uncertainties
• BB-full: Per-sample gamma params, per-sample Gaussian constraints (original behavior)

Tests validate:

• Combined uncertainties: sigma = sqrt(sum(errors^2))
• Poisson constraint: Poisson(tau | gamma * tau) where tau = (nu/sigma)^2
• Gaussian constraint: N(1.0 | gamma, relerr) where relerr = sigma/nu
• Shared gamma parameters correctly modify all sample rates
• Backward compatibility with BB-full mode

[codecov]

Codecov Report

❌ Patch coverage is 26.31579% with 42 lines in your changes missing coverage. Please review.
✅ Project coverage is 98.18%. Comparing base (66bb6f6) to head (7136718).
⚠️ Report is 5 commits behind head on main.
✅ All tests successful. No failed tests found.

Files with missing lines	Patch %	Lines
src/pyhs3/distributions/histfactory/__init__.py	25.00%	35 Missing and 4 partials ⚠️
src/pyhs3/distributions/histfactory/modifiers.py	40.00%	2 Missing and 1 partial ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##              main     #191      +/-   ##
===========================================
- Coverage   100.00%   98.18%   -1.82%     
===========================================
  Files           37       37              
  Lines         2257     2311      +54     
  Branches       234      249      +15     
===========================================
+ Hits          2257     2269      +12     
- Misses           0       37      +37     
- Partials         0        5       +5     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[Phmonski]

Parameters are also optional for the StatErrorModifier() of a Poisson constraint
parameters: list[str] -> parameters: list[str] | None = None

[kratsg]

hep-statistics-serialization-standard/hep-statistics-serialization-standard#68

I don’t think anything related to parameter naming should be optional. Makes it harder because then you can’t handle correlations

[Phmonski]

I found some already published HS3 JSON files on HEPData, where the Poisson staterror modifier boils down to:

{
    "constraint": "Poisson",
    "name": "staterror",
    "type": "staterror"
}

These workspaces raise errors during input validation if parameters are required.

[kratsg]

Yes, but they need to create a parameter of some name, and the name isn't dictated by the HS3 spec, and if it's left up to implementation, you're stuck trying to ascertain how to name the parameter associated with the staterror. It should be explicit.