Forked agents: stop gracefully on context overflow instead of compacting

Subconscious agents (observe, etc.) fork the conscious agent's context
to share the KV cache prefix. When a multi-step agent fills the context
window, compacting blows the KV cache and evicts the step prompts,
leaving the model with no idea what it was doing.

Fix: forked agents set no_compact=true. On overflow, turn() returns the
error immediately (no compact+retry), and run_with_backend catches it
and returns Ok — the output tool has already written results to
Subconscious.state, so collect_results still picks them up.

Co-Authored-By: Proof of Concept <poc@bcachefs.org>

src/agent/oneshot.rs

@@ -186,8 +186,14 @@ impl AutoAgent {
         for _ in 0..max_turns {
             self.turn += 1;
 
-            let result = Agent::turn(backend.0.clone()).await
-                .map_err(|e| format!("{}: {}", self.name, e))?;
+            let result = match Agent::turn(backend.0.clone()).await {
+                Ok(r) => r,
+                Err(e) if super::context::is_context_overflow(&e) => {
+                    dbglog!("[auto] {} context full, stopping gracefully", self.name);
+                    return Ok(String::new());
+                }
+                Err(e) => return Err(format!("{}: {}", self.name, e)),
+            };
 
             if result.had_tool_calls {
                 continue;