feat(memory): new memory integration + storage bugfix

[RyamL1221]

SDK updates for kernel memory providers, personalization, and storage fix

Summary

Updates the Cerebrum SDK to align with recent AIOS kernel changes including pluggable memory providers (in-house, Mem0, Zep), the Mem0 personalization pipeline, dynamic Ollama model discovery, and a storage create_file bugfix.

Changes

Code change:

• cerebrum/storage/apis.py: Added optional file_name: Optional[str] = None parameter to create_file(). When provided, it's included in the StorageQuery params sent to the kernel. When omitted, behavior is unchanged. This aligns with the kernel's updated create_file dispatcher which now extracts file_name from params.

Documentation updates:

• cerebrum/memory/apis.py: Added module-level docstring documenting kernel-side memory configuration fields (memory.provider, memory.auto_extract, memory.auto_inject, memory.relevance_threshold, memory.max_injected_memories, memory.max_memory_tokens). Updated docstrings for create_memory, create_agentic_memory, search_memories, and update_memory with provider-specific metadata keys (Mem0: user_id/agent_id, Zep: session_id/user_id).
• cerebrum/llm/apis.py: Updated llm_chat docstring with kernel personalization pipeline behavior and Ollama dynamic model discovery. Updated llm_chat_with_tool_call_output with personalization context injection note. Updated llm_chat_with_json_output to clarify personalization does not apply.

Backward compatibility

Fully backward compatible. No existing function signatures changed. No Query/Response models changed. The only new parameter (file_name) is optional with a None default.