added model selection framework

[Devguru-codes]

I have added ModelSelectionResult dataclass and compare_models() function in file osipy\dce\fitting.py that fits multiple DCE models to the same data and compare them using aic ,bic and R^2.
ModelSelectionResult class holds the best-model map, aic, bic, R^2 score maps, and the full DCEFitResult class which is Result container for DCE model fitting for each model.
compare_models() takes a list of model names, fits each one using the fit_model() pipeline and calculates aic, bic, and return which model is better at each voxel.
3 selection criteria for user - aic, bic and R^2 which is default criterion.
The code also has error handling for edge cases. Every warning is logged and even if 1 fails, other models continues to fit.
Feature fixes #101 is implemented here.
Updated PR also fixes #102

[Devguru-codes]

Since the feature implementation and bug fix was in same file so I fixed in same file. @ltorres6 sir please see it. This is just a proof of concept PR, I will update it after your suggestion accordingly. I closed the previous PR as its issue #102 was fixed in this.Thank you sir.

[RajdeepKushwaha5]

Hey! Nice work adding model comparison — that's a useful feature. A few suggestions from reviewing the diff:

1. except Exception: without as e: (issue #102 fix)

The silent pass is replaced with logging, which is great. However the pattern used here:

except Exception:
    logger.debug(...)

doesn't capture the exception object. The rest of the codebase (e.g., bayesian_ivim.py line 206) uses except Exception as e: and logs the actual error:

except Exception as e:
    logger.debug("Voxel %d fitting failed: %s", idx, e)

This gives much better diagnostic information when debugging fitting failures.

2. GPU compatibility in compare_models()

compare_models() uses np.asarray(concentration) directly. The osipy codebase is designed to be GPU/CPU agnostic using xp = get_array_module(array) (see osipy/common/backend/). Hardcoding np.asarray() would break when CuPy or JAX arrays are passed. Consider:

from osipy.common.backend import get_array_module
xp = get_array_module(concentration)

3. Bundling a bug fix (#102) with a feature (#101)

This PR bundles the #102 bug fix (4 lines) with the #101 model-selection feature (200+ lines). Separating them into individual PRs makes review easier and lets the quick bug fix land independently. The contributing guide recommends focused, single-concern PRs.

Hope this helps! Happy to discuss any of these points.

[ltorres6]

Closing. See discussion for reasoning.