Add visualize_representative_images for image-only topic models

[Gsebs]

What does this PR do?

Implements the approach discussed in #2236: a Plotly visualization that steps through each topic's representative-image collage (as created by VisualRepresentation) with a slider, at readable resolution — using only existing dependencies (plotly + numpy; no dash, no backend switch, no custom HTML/CSS), as preferred in that thread.

topic_model.visualize_representative_images()

The slider selects a topic; the title subtitle updates to the selected topic's label and the topic's collage is shown at its original resolution.

How it's structured

Follows the existing plotting conventions:

• standalone function in bertopic/plotting/_images.py, re-exported in bertopic.plotting
• thin delegating method BERTopic.visualize_representative_images() (placed after visualize_documents)
• API docs page (docs/api/plotting/representative_images.md) + mkdocs nav entry
• short usage note in the multimodal docs, next to the existing base64/HTML-table example
• supports topics= subsetting, custom_labels (bool or aspect name), and a configurable aspect name (default "Visual_Aspect")

Tests

tests/test_plotting/test_images.py mirrors the existing plotting-test idioms (parametrized over the fitted-model fixtures). Since the test extra doesn't include vision/PIL, tests inject numpy-array images into topic_aspects_ — the function consumes anything np.asarray accepts, so real PIL collages work identically.

Verified locally:

• full test suite green on unmodified master first (273 passed), then 70/70 plotting tests (63 existing + 7 new) with this change
• pre-commit (ruff format/check, whitespace, yaml) clean
• figure behavior verified in a real browser: initial trace visibility, slider steps, and per-step title/visibility updates

Happy to rename the function or adjust scope however you prefer, @MaartenGr — and I can attach a demo HTML if helpful.

Fixes #2236