[BUG] Visualization bugs in ShapeletClassifierVisualizer for multivariate data

[satwiksps]

Describe the bug

There are some critical issues in ShapeletClassifierVisualizer when working with multivariate time series (n_channels > 1):

1. Recursion Bug in Legends: The plotting function recursively appends channel information to the label string inside a loop (e.g., "Label channel 0 channel 1..."), resulting in illegible legends that obscure the plot.
2. Broadcasting Error: The method plot_distance_vector fails with a ValueError: operands could not be broadcast together for multivariate inputs because the normalization logic does not use keepdims=True.
3. Visual Clarity: The default styling for multivariate plots makes it difficult to distinguish between channels and classes (same colors used for different channels).

This issue was mentioned by @baraline during the discussion of documentation improvements here #3027 (reply in thread)

Steps/Code to reproduce the bug

import matplotlib.pyplot as plt
import numpy as np
from aeon.classification.shapelet_based import RDSTClassifier
from aeon.visualisation import ShapeletClassifierVisualizer

def test_legend_bug():
    X = np.random.rand(10, 3, 30) 
    y = np.array([0, 0, 0, 0, 0, 1, 1, 1, 1, 1])
    clf = RDSTClassifier(max_shapelets=5, random_state=42)
    clf.fit(X, y)
    viz = ShapeletClassifierVisualizer(clf)
    figs = viz.visualize_shapelets_one_class(X, y, class_id=0, n_shp=1)
    output_file = "legend_test_output.png"
    figs[0].savefig(output_file)

if __name__ == "__main__":
    test_legend_bug()

Expected results

• No Crash: The visualization should generate without a broadcasting ValueError.
• Clean Legend: The legend should display concise labels (e.g., "Class 0 channel 0") without recursive repetition.
• Clear Distinction: Multivariate plots should use distinct visual cues (e.g., color for channels, line style for classes) to be readable.
• Proper Alignment: No text overrun or overlap should be there.

Actual results

Crash: ValueError: operands could not be broadcast together with shapes...

Visuals (if crash fixed): The legend text grows exponentially with the number of channels, and lines are indistinguishable.

Versions

Details

System: python: 3.11.0 (main, Oct 24 2022, 18:26:48) [MSC v.1933 64 bit (AMD64)] executable: C:\Users\satwi\AppData\Local\Programs\Python\Python311\python.exe machine: Windows-10-10.0.26200-SP0 Python dependencies: aeon: 1.3.0 pip: 22.3 setuptools: 65.5.0 scikit-learn: 1.7.2 numpy: 2.2.6 numba: 0.61.2 scipy: 1.15.3 pandas: 2.3.3