Provide an equivalent text description for the main figure

## Problem

The README's primary TableVerse overview is a complex informational figure, but its only text alternative is `TableVerse main figure`.

The image communicates several pieces of information that are not available from that label alone:

- many examples of reconstructed tabletop scenes and layouts;
- the TableVerse-100K identity;
- the headline scale of 100K scenes, 1M objects, and 35K categories;
- word clouds illustrating object labels and spatial/color/size attributes.

The default-branch and web-branch READMEs both use the same image and similarly short alternative text. The surrounding README contains no equivalent prose description.

## Reproduction

1. Read `README.md` with images disabled or with a screen reader.
2. Navigate to `docs/assets/main.png`.
3. Observe that only the phrase `TableVerse main figure` is available in place of the image's statistics and visual summary.

## Expected behavior

A complex informational figure should have a concise alt label plus an adjacent structured description of its meaningful content and headline statistics.

## Actual behavior

Readers who cannot perceive the image receive only its generic name.

## Impact

The repository's primary dataset overview and scale summary are unavailable to screen-reader and text-only users.

## Why this report has no PR

The same figure is referenced by two orphan branches, and an equivalent description should reflect which visual takeaways the paper authors consider authoritative—especially whether the displayed 1M/35K figures should retain the paper's `nearly`/`over` qualifiers. Once that wording is confirmed, both READMEs and the project-page image description can be updated together without introducing a new metadata inconsistency.

## Evidence

- Default branch: `main@2de83cb43b759c7b3d07c749c6a3efccd4528007`, `README.md` line 8
- Pages source branch: `web@66efcf4f1f8a4aaf5cc937e5b6d5bd9724f3ea04`
- Shared image blob: `f2a8df9699fb4e740e535a9976ea831f542150bc`

Start by comparing README.md on the default branch with the web branch README, then inspect docs/assets/main.png and the existing `TableVerse main figure` alternative text. Confirm with the paper authors whether the 1M and 35K figures should retain their qualifiers. Done means adding an equivalent structured description to both READMEs and the project-page image description without creating inconsistent metadata.

Written by the indexing model from the issue text.