feat: add Atlas Cloud WRBench first-frame provider

[binyangzhu000-sudo]

Summary

Adds Atlas Cloud as an optional WRBench first-frame image provider. The implementation follows Atlas Cloud's asynchronous image flow: one generation submission, bounded prediction polling, then output download.

User-Visible Behavior

wrbench firstframe --provider atlascloud can generate Natural-25 first frames with an Atlas Cloud image model. Existing bundled frames and providers are unchanged.

Affected Pipeline / Benchmark

• Pipeline(s): WRBench first-frame generation
• Benchmark(s): WRBench Natural-25
• Runtime profile(s): N/A
• Public entrypoint(s): wrbench firstframe

Change Type

• Bug fix
• Model integration
• Benchmark integration
• Pipeline/runtime change
• Documentation only
• Test/QA tooling

Asset / API / GPU Requirements

• Downloads or large assets: None
• API keys or quota: Atlas Cloud API key and image-generation quota for live use
• GPU / simulator requirements: None
• Official repository or checkpoint assumptions: None

Commands Run

Command	Result	Notes
python3 -m pytest -q worldfoundry/evaluation/tasks/execution/runners/wrbench/test_atlas_cloud_firstframe.py worldfoundry/evaluation/tasks/execution/runners/wrbench/test_wrbench_generation_bridge.py	Pass	4 tests passed
python3 -m ruff check --select E4,E7,E9,F --ignore F401 ...	Pass	Changed provider and test files
python3 -m py_compile ...	Pass	Changed Python files
bash scripts/docs/build.sh --skip-bootstrap	Pass	1,257 static pages generated
Live Atlas Cloud smoke	Pass	Seedream v5.0 Lite; one POST; completed image downloaded

Checkpoint / API Key Needs

• Checkpoints / weights: None
• Required environment variables: ATLASCLOUD_API_KEY (passed to --api-key)
• API providers: Atlas Cloud
• Local cache assumptions: None

Sample Artifact Evidence

• Artifact path or link: Local smoke output generated during validation; not committed
• What it demonstrates: Submission, prediction polling, URL extraction, and image download complete successfully
• Known limitations: Atlas Cloud currently returns one output per first-frame request, so --n must be 1

Validation Matrix Status

Area	Status	Evidence / Command
Unit or focused regression	Pass	4 focused tests
Real inference validation	Pass	Live Seedream v5.0 Lite generation
Streaming or multi-turn path	N/A	Image generation is asynchronous polling
Benchmark runner / metric path	Pass	Existing WRBench generation bridge tests
Docs build or link check	Pass	Fumadocs production build
GPU/API-key dependent path	Pass	Live API-key smoke; no GPU required

Leaderboard Validity Impact

• Readiness status changed: No
• leaderboard_valid impact: None
• Scorecard/preflight evidence: N/A
• Demo or contract-only limitations: This only adds an optional first-frame source and does not change scoring

Compatibility And Risk

• Backward compatibility impact: None; provider selection remains explicit
• Expected resource cost: One billed image generation per requested frame
• Failure modes or rollout concerns: Missing credentials, rejected model parameters, provider failure, or prediction timeout surfaces as an exception; generation POSTs are not retried

Checklist

• I kept the change narrowly scoped.
• I did not commit secrets, API keys, large checkpoints, or generated cache files.
• I reused fixtures from data/test_cases/ or assets from data/benchmarks/ where practical.
• I updated docs when public behavior, install steps, or entry commands changed.
• I documented API/GPU/checkpoint/official-repo requirements or confirmed none are needed.
• I did not promote demo, contract-only, normalizer-only, API-blocked, or partial evidence to leaderboard validity.