Add ScatterShift augmentation (forward EMSC model)

[prabeshjoshi]

Summary

• Adds chemotools.augmentation.ScatterShift, a transformer that injects random
  multiplicative scatter and a random polynomial baseline using the forward
  Extended Multiplicative Scatter Correction (EMSC) model:
  x_aug = m·x + Σ cᵢ·λⁱ.

Why is this change needed?

• The augmentation module covers additive noise (AddNoise), a constant baseline
  (BaselineShift), a scalar multiplicative factor (SpectrumScale), axis shifts,
  and peak broadening, but has no augmenter for multiplicative scatter combined
  with a wavelength-dependent baseline: the dominant physical scatter signature
  in NIR/vibrational spectra. ScatterShift fills that gap and is the forward
  counterpart of the existing ExtendedMultiplicativeScatterCorrection: augment
  with ScatterShift, correct with EMSC.

Closes

• Closes Feature: scatter augmentation via forward EMSC model #292

Related issues

Type of change

• New feature
• Tests only

What changed?

• Added chemotools/augmentation/_scatter_shift.py implementing ScatterShift.
• Registered ScatterShift in chemotools/augmentation/__init__.py (import + __all__).
• Added tests/augmentation/test_scatter_shift.py.

What did NOT change?

• No changes to chemotools.scatter (EMSC/MSC correctors are unchanged).
• No changes to any other augmentation transformer.
• No docs page updated yet (see Documentation notes below).

API and compatibility impact

• Public API added

Notes

• scikit-learn API compatibility: ScatterShift inherits DocLinkMixin, TransformerMixin, OneToOneFeatureMixin, BaseEstimator, uses
  _parameter_constraints for all four parameters, and passes
  check_estimator() at default parameters (identity transform when
  multiplicative_scale=0.0 and additive_scale=0.0).
• Affected modules/classes/functions: chemotools.augmentation (new class
  ScatterShift); chemotools/augmentation/__init__.py (new export).
• Backward compatibility considerations: purely additive change, no existing
  public API modified.

Validation

Validation notes

• Verified locally (outside task) via ruff format --check, ruff check, and
  python -m pytest tests/augmentation/test_scatter_shift.py -v : all passing,
  6/6 tests, no lint or format issues. Have not yet run the full task check /
  task test:matrix suite; deferring to CI for the dependency-floor and
  multi-version matrix.

Tests

• Added new tests

Test coverage details

• Relevant test files: tests/augmentation/test_scatter_shift.py
• Edge cases covered: identity behavior at default parameters, multiplicative-only
  case (order=0), polynomial-baseline case (order≥1), reproducibility under a
  fixed random_state, and an explicit check that the internal polynomial basis
  is identical to ExtendedMultiplicativeScatterCorrection's design matrix.
• Numerical / estimator behavior checked: check_estimator() compliance;
  round-trip sanity (augmenting then EMSC-correcting recovers the original signal),
  checked manually, not asserted as a formal test.

Documentation

• Docs updated
• Docstrings updated

Documentation notes

• ScatterShift has a full numpydoc docstring (model equation, EMSC
  cross-reference, Afseth & Kohler 2012 reference, runnable example), but the
  augmentation methods page (methods/augmentation_methods.html) is not yet
  updated. Happy to add an entry if you point me to the right file/format.

Dependency / build / CI impact

• No dependency changes

Reviewer focus

Please focus on:

• Whether the forward-model framing (and reuse of the EMSC polynomial basis) is
  the right approach, versus a different parameterization.

Checklist

• I linked the relevant issue(s) or explained why none exists
• I kept this PR scoped to a single purpose
• I added or updated tests where appropriate
• I updated documentation where appropriate
• I verified the change locally using the relevant tasks above
• I considered backward compatibility and public API impact
• I am ready for review