feat: support model fallback endpoints - #1083

Summary

This PR adds framework-level support for model fallback endpoint configuration.

It introduces ModelFallbackEndpoint so Agent(model_fallbacks=...) can accept fallback models with independent provider, API base, API key, and LiteLLM extra config. It also exports the new type from veadk, wires fallback configs into RetryingLiteLlm, and adds runtime compatibility warnings for runtimes that do not support explicit model fallbacks.

Changes

• Add ModelFallbackEndpoint and ModelFallbackConfig.
• Support same-provider string fallbacks and cross-provider endpoint fallbacks in Agent.
• Export ModelFallbackEndpoint from veadk.__init__.
• Pass structured fallback configs through RetryingLiteLlm.
• Warn when external runtimes drop Agent(model_fallbacks=...).
• Add unit coverage for fallback model behavior and runtime compatibility messaging.

Testing

• uv run pre-commit run --files veadk/__init__.py veadk/agent.py veadk/models/retrying_lite_llm.py veadk/runtime/compat.py tests/test_agent.py tests/runtime/differential/test_runtime_parity.py
• uv run pytest tests/test_agent.py tests/runtime/differential/test_runtime_parity.py -q

Result: 44 passed, 2 xfailed