Refactor: Relocate event ID aliasing logic out of cognitive_engine (Resolve Responsibility Leakage)

[narv-lab]

Description

In our system architecture, the relationship between cognitive_engine and memory closely mirrors that of a user-space application and the OS kernel's memory management.
Currently, we convert absolute event_ids (UUIDs) into short aliases (e.g., REF-1) before passing them to the LLM. This process is analogous to an OS Memory Management Unit (MMU) translating Physical Addresses into Virtual Addresses to provide a limited, safe context window for the application.

The Problem (Responsibility Leakage)

Presently, the cognitive_engine (the "application") holds the alias_map and performs the translation (aliasing and de-aliasing) itself.
In a proper OS architecture, a user-space application should never manage its own page tables or perform physical-to-virtual address translations. Having this logic inside cognitive_engine constitutes a clear responsibility leakage, tightly coupling the pure inference pipeline with memory address resolution mechanics.

Proposed Solution

We need to relocate the alias mapping responsibility outside of the cognitive_engine to act as a proper MMU.

1. Relocate Mapping Logic: Move the alias_map generation and the UUID ↔ Alias translation logic into the kernel_mediator (or the memory facade).
2. Transparent Input/Output: The cognitive_engine should receive context data that is already mapped to aliases (Virtual Addresses). It should output its causal_links using those aliases without knowing the underlying UUIDs.
3. Intercept and Translate: The kernel_mediator will intercept the LLM's output, translate the aliases back to the real UUIDs (Physical Addresses), and pass them to the memory module for storage.

Benefits

• Enforces strict Separation of Concerns (SoC).
• cognitive_engine becomes a purer, stateless inference pipeline.