Stop tokens hemorragy

[Malakof]

We need a way to point archon agents to a pre-computed environment, that include pre-processed llms sessions so the massive amount of tokens required for research are available through sessions fork to all agents that will work on related issues / features. This is most obvious when agents write multiple amends through several passes of an llm as a judge.

This is mandatory to deploy archon at scale, otherwise it's a token sink.

This is also mandatory for the planet and no responsible AI engineer should contribute to token waste.