sync main to dev

[e06084]

No description provided.

[gemini-code-assist]

Code Review

This pull request ensures configuration isolation by deep copying dynamic_config when setting rules or LLM configurations, and adds corresponding unit tests. Feedback was provided to avoid using model_dump() when iterating over configuration arguments, as it converts nested Pydantic models into dictionaries and leads to attribute access errors; using dict().items() is recommended instead to preserve object types.

[gemini-code-assist]

Using model_dump() on a Pydantic model converts all nested models into dictionaries. When these dictionaries are later assigned back to the configuration object via setattr, the nested fields lose their Pydantic model type and become plain dicts. This will cause an AttributeError when the code attempts to access attributes on those nested objects (e.g., config.embedding_config.model). To preserve the object types, iterate over the model instance directly or use dict(model).items() which performs a shallow dump.

Suggested change

	config_default = rule.dynamic_config.model_copy(deep=True)
	# Iterate over rule_config fields using Pydantic's model_dump()
	for k, v in rule_config.model_dump().items():
	if v is not None:
	config_default = rule.dynamic_config.model_copy(deep=True)
	# Iterate over rule_config fields directly to preserve nested model types
	for k, v in dict(rule_config).items():
	if v is not None:

[gemini-code-assist]

Same as the issue in set_config_rule. EvaluatorLLMArgs contains embedding_config, which is a nested Pydantic model. Using model_dump() will turn it into a dictionary, breaking subsequent attribute access on the configuration object. Iterating over dict(llm_config).items() is safer as it preserves nested objects and correctly handles extra fields allowed by the model configuration.

Suggested change

	config_default = llm.dynamic_config.model_copy(deep=True)
	# Iterate over llm_config fields using Pydantic's model_dump()
	for k, v in llm_config.model_dump().items():
	if v is not None:
	config_default = llm.dynamic_config.model_copy(deep=True)
	# Iterate over llm_config fields directly to preserve nested model types
	for k, v in dict(llm_config).items():
	if v is not None: