Add "tokens" parameter for entity-controlled context size in memory retrieval

[rob-mosher]

Problem: Entities need control over HOW MUCH memory context they receive, not just WHICH topics.

Proposed Solution: Add tokens parameter specifying desired context window size:

python

relational_state.request(
  topic="foobar",
  tokens=5000,
  exploration=0.3
)

Why This Matters:

• Cost control: Entities manage their own token budgets
• Precision vs. breadth: Small windows for focused work, large for deep context
• Autonomy: Entity decides tradeoff between detail and scope

Implementation:

• Interacts with RLM compression (retrieve more, compress to fit)
• Works WITH exploration parameter (exploration finds relevant memories, tokens controls how much to return)

---

Collaborator: Rob Mosher | Direct | Human | Proposed tokens parameter for entity autonomy
Collaborator: Claude Sonnet 4.5 | Direct | AI | Drafted issue