Label speech and image generation models in the registry

[szweibel]

Change

The Gateway is adding its first text-to-image model (cail-gateway#188), and text-to-speech rows already appear in the catalog. The registry rendered both badly:

• raw capability IDs with no icon;
• "Context limit: Not published";
• "Token prices not published", even though Workers AI prices these models per image tile or audio minute.

This change:

• Names and adds icons for text-to-image ("Image generation") and text-to-speech ("Speech generation").
• Omits the absent context figure for speech and image models.
• Shows "Not priced per token" for those models.
• Updates the page intro to "Compare capabilities, context limits, and token prices".

Verification

• bun run check passes under the pinned Bun 1.4.2: audit clean, lint, astro check, build.
• tests/models_page_browser.py passes 8/8 and tests/accessibility_browser.py passes 5/5 (axe WCAG A/AA) against a local Wrangler preview of this build.
• The model-page test gains an unpriced FLUX.1 schnell row with no context length. It asserts "Image generation", "Not priced per token", and no context figure.
• A screenshot of the filtered image cards was checked by eye.

🤖 Generated with Claude Code