feat: initial commit with KernelPLS implementation and tests

[inarteroger]

Why is this change needed?

Adds a KernelPLS regressor to the chemotools.models module.
KernelPLS extends classical PLS to nonlinear relationships by applying
kernel methods and operating in reproducing kernel Hilbert space (RKHS).

This provides a flexible alternative to standard PLSRegression,
particularly useful for nonlinear spectral data.

---

Type of change

• New feature

---

Description

Implementation of KernelPLS following the method described in:

[1] Rosipal, R. & Trejo, L. J. (2001).
Kernel Partial Least Squares Regression in Reproducing Kernel
Hilbert Space. Journal of Machine Learning Research, 2, 97–123.
http://www.jmlr.org/papers/volume2/rosipal01a/rosipal01a.pdf

Implementation of Kernel Partial Least Squares (KernelPLS),
supporting multiple kernel functions (linear, rbf, polynomial, sigmoid)
and compatible with the scikit-learn estimator API.

The model computes pairwise kernels, centers them using KernelCenterer,
and applies PLSRegression on the transformed feature space.

---

What was added

• KernelPLS estimator in chemotools/models/_kernel_pls.py
• Full sklearn API compliance:
  • fit, predict, transform, fit_transform, score
• Support for multiple kernels:
  • linear, rbf, poly, sigmoid

---

Tests added

• sklearn estimator compliance (check_estimator)
• fit/predict/transform correctness
• consistency with sklearn reference implementation:
  (KernelCenterer + PLSRegression)
• input validation
• pipeline and GridSearchCV compatibility