feat: add Meta Model API and Muse Spark

[neptune-me]

Summary

• add Meta Model API as a first-class provider with Muse Spark model discovery
• discover supported Muse Spark versions from the Meta model catalog, including cached results
• map the provider to the existing OpenAI-compatible backend
• adapt CAMEL Chat Completions tool schemas for Meta by omitting strict mode while preserving dictionary schemas
• add focused provider, alias, model-discovery, production adapter-wiring, and tool-schema regression coverage

Verification

• uv run --directory backend pytest tests/app/model tests/app/controller tests/app/agent/test_agent_model.py tests/app/component/test_model_validation.py -q (315 passed, 3 skipped)
• vitest run test/unit/lib/llm.test.ts test/unit/lib/providerModels.test.ts --no-cache (4 passed)
• npm run check:i18n
• npm run type-check
• focused ESLint, Ruff, Prettier, SVG, design-token, Bandit, and whitespace checks
• live Muse Spark 1.3 validation on macOS 26.6.1 arm64; requests completed with finish_reason=stop

Notes

Meta follows the same empty initial model selection used by the providers in #1823 and #1646. Refreshing the catalog exposes Muse Spark 1.1, 1.2, 1.3, contributor variants, and future muse-spark-* versions. The Meta-specific tool compatibility hook intentionally targets direct api.meta.ai Chat Completions requests; proxied endpoints and Responses use separate compatibility paths. No credentials or local environment files are included.

Closes #1894