Add Atlas Cloud image generation skill

[binyangzhu000-sudo]

Summary

• add an optional Atlas Cloud image generation Skill with a standard-library CLI
• validate local inputs, submit exactly one generation POST, poll prediction GETs with bounded retries, and download the first HTTPS output
• add offline coverage and UI metadata, and bump the Plugin patch version to 1.2.1

Validation

• python3 -m compileall -q skills/atlas-image-skill
• uvx ruff check skills/atlas-image-skill
• python3 -m unittest discover -s skills/atlas-image-skill/tests -v (8 tests)
• python3 ~/.codex/skills/.system/skill-creator/scripts/quick_validate.py skills/atlas-image-skill
• python3 ~/.codex/skills/.system/plugin-creator/scripts/validate_plugin.py .
• bash scripts/test-plugin-install.sh
• python3 skills/atlas-image-skill/atlas_image.py --help
• git diff --check

Integration check

The live schema for google/nano-banana-2-lite/text-to-image-developer was verified before implementation. One generation POST was attempted and was not retried; the edge returned HTTP 403 for the default Python User-Agent before creating a prediction. Read-only diagnostics confirmed that an explicit client User-Agent reaches the API, where the available credential currently returns HTTP 402 (insufficient balance). The helper now sends a stable User-Agent, and a second paid POST was intentionally not attempted.

Compatibility and security

• the Skill is separate and optional; existing image providers and defaults are unchanged
• no third-party Python dependencies or additional browser/runtime permissions are introduced
• credentials are read only from ATLASCLOUD_API_KEY or the compatibility alias ATLAS_CLOUD_API_KEY