Add context compaction when approaching token limit

Currently agent has a 500k token guardrail that kills the run. Instead: 1. When approaching limit (e.g. 80%), compact the conversation history 2. Summarize older messages, keep recent ones verbatim 3. Continue the run with compacted context

This enables long-running agents that can work on big tasks.

Alternative: just remove the limit like we did for iterations/cost. But compaction is more useful long-term for efficiency.

References:

• Omni/Agent/Engine.hs:642 - guardrailMaxTokens = 500000
• Omni/Agent/Engine.hs:1077 - already has 0 = no limit logic