fix: squeeze regressor scores before scalar assignment for numpy 2.x compat

[SudipSinha]

Summary

In numpy 2.x, PyTorch model output for single-output regressors has shape (n, 1) rather than (n,), so model_scores[i] is an array of shape (1,) rather than a scalar. Assigning a sequence to a scalar element of a numpy array raises ValueError in numpy 2.x:

ValueError: setting an array element with a sequence.
TypeError: only 0-dimensional arrays can be converted to Python scalars

Fix: np.squeeze collapses the size-1 dimension; float() converts the resulting 0-d array to a Python scalar before the assignment. This is a no-op for backends (e.g. sklearn) that already return a scalar from predict.

File: dice_ml/explainer_interfaces/explainer_base.py:813

# before
model_output[i] = model_scores[i]

# after
model_output[i] = float(np.squeeze(model_scores[i]))

Impact

Fixes 14 failing regression tests across dice_random, dice_KD, and dice_genetic with the PyTorch backend. No change in numeric output — np.squeeze + float extracts the identical scalar value; the model_output array dtype (float32) still governs precision.

Test plan

• All 4 TestDiceGeneticRegressionMethods PYT tests now pass (were ValueError)
• Full suite: 267 passed, 13 failed (down from 253 passed, 27 failed) — remaining 13 failures are a separate pre-existing pd.read_json issue unrelated to this change
• No regressions in classifier path or sklearn regressor path

Tested on Python 3.12, 3.13, 3.14 with numpy 2.5.3.

[SudipSinha]

Related PRs for context:

• fix: introduce explicit conversion to scalar to avoid implicit conversion issues after numpy 2.0.0 #475 — fixes numpy 2.x compatibility in the classifier paths of explainer_base.py and dice_genetic.py (int() conversions via .item()). Complementary to this PR; neither duplicates the other.
• Add python 313 to build tests #457 — adds Python 3.13 to the CI matrix (currently on hold pending TensorFlow 3.13 support).

[SudipSinha]

Fixes #474 (regressor path of the numpy 2.x bug).

[SudipSinha]

Hi @amit-sharma and @gaugup, can you have a look at this one-line change PR? TY!