Add data-driven background param models: ABCD, SmoothABCD, ExtendedABCD, SmoothExtendedABCD

[davidwalter2]

• Rename poi_model → param_model and add explicit npoi/nnui split so models can declare pure nuisances with npoi=0
• Support multiple --paramModel invocations: each call appends a spec; load_models() wraps them in CompositeParamModel automatically
• ABCD: data-driven background estimation using 4 regions. Free per-bin parameters a, b, c for regions A, B, C; D predicted as D = A·C/B · mc_factor_D. npoi=0,
  nnui=3·n_bins
• SmoothABCD: like ABCD but all three free-parameter regions are parameterised with an exponential polynomial along one nominated axis: val(x) = exp(p₀ + p₁·x̃ + …). Reduces
  parameters from 3·n_bins to 3·n_outer·(order+1). Default order=1 (log-linear)
• ExtendedABCD: 6-region variant adding extra sideband bins Ax, Bx. Log-linear fake rate extrapolation in the x direction: D = C·Ax·B² / (Bx·A²) · mc_factor_D. npoi=0,
  nnui=5·n_bins
• SmoothExtendedABCD: combines smooth parameterisation with the extended 6-region formula. All five free-parameter regions (A, B, C, Ax, Bx) use exponential polynomials.
  npoi=0, nnui=5·n_outer·(order+1)

All four models share the same design principles: positivity guaranteed (squaring for ABCD/ExtendedABCD, exp for smooth variants), MC correction factor absorbs non-unity MC
templates, and they compose freely with Mu or other models via CompositeParamModel.

CLI examples

# Standard ABCD
--paramModel Mu --paramModel ABCD nonprompt ch_A ch_B ch_C ch_D
                                                               
# Log-linear smooth fake rate along pt                                                                                                                                                
--paramModel Mu --paramModel SmoothABCD pt nonprompt ch_A ch_B ch_C ch_D
                                                                                                                                                                                      
# 6-region extended ABCD
--paramModel Mu --paramModel ExtendedABCD nonprompt ch_Ax ch_Bx ch_A ch_B ch_C ch_D
                                                                                   
# Smooth 6-region, quadratic polynomial                                                                                                                                               
--paramModel Mu --paramModel SmoothExtendedABCD pt order:2 nonprompt ch_Ax ch_Bx ch_A ch_B ch_C ch_D                                                                        

Test plan

• tests/make_abcd_tensor.py and tests/make_extended_abcd_tensor.py produce valid tensors
• All four models load with correct npoi/nnui/nparams
• Fits converge with --saveHists for ABCD, SmoothABCD, ExtendedABCD, SmoothExtendedABCD
• New CI jobs make-abcd-tensors and abcd-fits cover all four models