Add provider binding validation and live-chat agent config

[ManuelKugelmann]

Summary

This PR adds comprehensive validation to ensure all agents have properly configured providers and models, with a new test suite that verifies provider bindings against the librechat configuration. It also explicitly configures the live-chat agent to use Anthropic's Claude Sonnet 4.6 model.

Key Changes

• New test class TestProviderBinding: Validates that every agent has an effective provider and model by resolving overrides from agent-models.json against base agents.json configuration
• Provider validation logic: Ensures all agent providers either:
  • Are native providers (OpenAI, Anthropic) requiring only .env keys
  • Exist as custom endpoints in librechat-user.yaml
• Live-chat agent configuration: Explicitly assigns the main chat agent to Anthropic's Claude Sonnet 4.6 model with a safeguard test to prevent downgrading to cheaper providers
• GitHub Models PAT documentation: Updated comment to clarify that classic PATs or fine-grained tokens with "Models" read permission are both acceptable

Implementation Details

• The effective_assignments fixture resolves the provider/model hierarchy: base agent config → layer defaults → per-agent overrides
• The NATIVE_PROVIDERS constant identifies providers that don't require YAML endpoint configuration
• The test_live_chat_not_downgraded test acts as a safeguard to prevent accidental cost-cutting that would degrade the primary chat experience
• All validation tests skip gracefully if librechat-user.yaml is not present (e.g., in minimal test environments)

https://claude.ai/code/session_01CYT2ttbEzMGq2uoZgRHo2X