Add parameter definitions for centroid-detection algorithms

Introduce `params.py` under `source/core/` to centralize algorithm parameter configurations (SLIC, Felzenszwalb, Quickshift, FBM, CCL, and Bessel). Implement frozen dataclasses to ensure immutability and reusability across datasets and algorithms.

Author: greg-ogs

source/core/params.py

@@ -0,0 +1,82 @@
+"""Algorithm parameters for the centroid-detection methods.
+
+Frozen dataclasses so the call sites can pass a single value object instead
+of a long, easy-to-mismatch keyword list, and so per-dataset configurations
+can be defined once and reused across algorithms.
+
+All defaults below are the fine-tuned values from the pre-refactor code at
+commit ``ab17d7b`` (the laser-spot dataset under ``images/``). When changing
+any default, also update the regression range in ``tests/test_smoke.py``.
+
+Notes on the Gaussian-blur sigmas (FBM + Bessel)
+------------------------------------------------
+The original code called ``cv2.GaussianBlur(img, (5, 5), 0)`` which lets
+OpenCV auto-compute sigma from the kernel size:
+
+    sigma = 0.3 * ((ksize - 1) * 0.5 - 1) + 0.8
+          = 0.3 * 1 + 0.8 = 1.1   for ksize = 5
+
+We now go through :func:`source.core.gpu_utils.gaussian_blur` (which uses
+``scipy.ndimage.gaussian_filter`` on CPU and ``cupyx.scipy.ndimage`` on GPU),
+so sigma is passed explicitly. ``blur_sigma = 1.1`` preserves the original
+smoothing strength.
+"""
+from __future__ import annotations
+
+from dataclasses import dataclass
+
+
+@dataclass(frozen=True)
+class SLICParams:
+    n_segments: int = 125
+    compactness: float = 10.0
+    sigma: float = 5.0
+
+
+@dataclass(frozen=True)
+class FelzenszwalbParams:
+    scale: float = 200.0
+    sigma: float = 0.5
+    min_size: int = 150
+
+
+@dataclass(frozen=True)
+class QuickshiftParams:
+    kernel_size: int = 21
+    max_dist: float = 50.0
+    ratio: float = 5.0
+
+
+@dataclass(frozen=True)
+class FBMParams:
+    """OpenCV-moments centroid: blur, threshold, morph close."""
+    blur_sigma: float = 1.1     # matches cv2.GaussianBlur(_, (5, 5), 0) default
+    threshold: int = 100
+    threshold_max: int = 255    # uint8 binary mask ceiling
+    morph_kernel: int = 5
+
+
+@dataclass(frozen=True)
+class CCLParams:
+    """scikit-image connected-components centroid."""
+    blur_sigma: float = 2.0
+    morph_disk: int = 5
+
+
+@dataclass(frozen=True)
+class BesselParams:
+    """Airy-disk (Bessel) fit.
+
+    ``window_size`` is the half-width of the ROI extracted around the
+    initial peak guess. ``initial_blur_sigma`` smooths the image before
+    locating that peak. The ``fit_*`` fields seed ``scipy.optimize.curve_fit``:
+
+      p0      = [A0,    x0_init, y0_init, fit_sigma_init, offset_init]
+      bounds  = [[0,    x_min,   y_min,   fit_sigma_min,  0           ],
+                 [inf,  x_max,   y_max,   inf,            offset_max   ]]
+    """
+    window_size: int = 80            # half-width of the ROI around the peak
+    initial_blur_sigma: float = 1.1  # matches cv2.GaussianBlur(_, (5, 5), 0)
+    fit_sigma_init: float = 5.0      # initial guess for the Airy sigma
+    fit_sigma_min: float = 0.1       # lower bound for the Airy sigma
+    fit_offset_max: float = 1.0      # upper bound for the background offset