feat(google): add gemini-3.1-flash-lite-image and fix Gemini image catalog errors

[Kamilbenkirane]

Summary

• Adds gemini-3.1-flash-lite-image (Nano Banana 2 Lite) to the Google image catalog: 1K-only resolution, 14 object reference images, generate+edit, minimal/high thinking levels.
• Auditing the sibling entries while verifying this one turned up real, pre-existing errors, fixed here:
  • gemini-3-pro-image-preview and gemini-3.1-flash-image-preview are deprecated and shut down June 25, 2026 — renamed to their GA replacement IDs gemini-3-pro-image and gemini-3.1-flash-image.
  • THINKING_LEVEL was ["minimal", "High"] everywhere in the Gemini 3.x image family; the API expects lowercase "high" (confirmed against an actual code sample in Google's docs). Fixed on all three Gemini 3.x entries.
  • gemini-3.1-flash-image (GA) gained a 512px quality tier over the preview version — added.
  • imagen-4.0-fast-generate-001 had a QUALITY: ["1K"] constraint, but Google's docs state the imageSize parameter is only supported on the Standard and Ultra Imagen 4 variants, not Fast — removed.
• Grepped the repo for the old preview IDs — no other code/tests referenced them, so the rename is safe.
• gemini-2.5-flash-image's aspect ratio list and reference-image count could not be independently confirmed against official docs and were left unchanged.

Test plan

• make ci passes (lint, format, mypy, bandit, tests with coverage)
• Manually loaded each changed/new Model entry and confirmed it validates
• Verified every changed value against Google's official Gemini API docs (model pages, image-generation guide, changelog/deprecations)

[claude]

Code review

1 issue found.

Bug: Potential invalid imageSize value "512px" on gemini-3.1-flash-image

celeste-python/src/celeste/modalities/images/providers/google/models.py

Lines 119 to 125 in 0799298

	"21:9",
	]
	),
	ImageParameter.QUALITY: Choice(options=["512px", "1K", "2K", "4K"]),
	ImageParameter.REFERENCE_IMAGES: ImagesConstraint(max_count=14),
	ImageParameter.THINKING_LEVEL: Choice(options=["minimal", "high"]),
	},

The QUALITY value is passed verbatim to the Gemini API as generationConfig.imageConfig.imageSize with no normalization (traced through GeminiQualityMapper → ImageSizeMapper → Choice.__call__). Every other model in this file uses the {n}K format ("1K", "2K", "4K"). "512px" is inconsistent with this format and may not be a valid imageSize enum token — if so, any request selecting this quality tier will be rejected by the Gemini API.

Please verify the exact string the Gemini API expects for this tier against the official Google documentation. If the correct value differs (e.g. "512" or another format), it should be updated here.