fix(ai): Resolve local inference 4096 n_ctx Exhaustion During Session Summarization

Description

The mcp_neo-mjs-memory-core_summarize_sessions tool routinely crashes internally with the local inference daemon error: The number of tokens to keep from the initial prompt is greater than the context length (n_keep: 5382 >= n_ctx: 4096).

This indicates that as Neo.ai.Agent reasoning loops expand during heavily contextual RLAIF sessions (due to increased swarm intelligence and tool use), the final session summarization pipeline strictly drops the payload and fails to index the session.

Architectural Rationale

• The underlying backend (e.g., MLX / openAiCompatible) currently defaults to a context limit of 4096 tokens when hosting gemma4:31b or similar local models.
• We must either explicitly bump this n_ctx limit at the deployment daemon layer (e.g., updating Python boot args to 8192), or we must instruct the memory-core MCP Server to truncate/chunk the combined prompt + thought + response strings before throwing the batch at the extraction endpoint.
• A2A Context: Failure to resolve this natively prevents the most complex agent problem-solving sessions from being squashed into Vector Summaries. Consequently, the Swarm loses immediate visibility to precisely the problems it struggled with the most.

Origin Session ID: 95bf4a2b-d84e-4f70-945b-f558ba924d3a