Fix context references dropped due to wrong isinstance check

[jnMetaCode]

Summary

In generate_chatml_messages_with_context(), document references from prior conversation turns are never included in the prompt because of an incorrect isinstance(item, dict) type check.

Bug

chat.context is List[Context] where Context is a Pydantic BaseModel. After ChatMessageModel.model_validate() in the Conversation.messages property, context items are Context objects, not raw dicts. The isinstance(item, dict) guard therefore filters out all items, producing an empty references string every time.

references = "\n\n".join(
    {
        f"# URI: {item.uri or item.file}\n## {item.compiled}\n"
        for item in chat.context or []
        if isinstance(item, dict)   # <-- always False for Context objects
    }
)

Fix

Remove the isinstance(item, dict) guard. The list is already typed as List[Context] and validated by Pydantic, so the type check is unnecessary and harmful.

Impact

Without this fix, the LLM does not see document references from earlier turns in multi-turn conversations, which can degrade response quality when prior context is relevant.