Incremental memory scoring with per-score persistence

score_memories_incremental now takes an async callback that fires
after each memory is scored. The callback:
- Writes the score to the conversation entry via set_score()
- Persists to memory-scores.json immediately
- Notifies the UI so the context screen updates live

Scoring no longer batches — each score is visible and persisted
as it completes. Does not touch the memory store.

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

src/subconscious/learn.rs

@@ -299,13 +299,18 @@ pub async fn score_memory(
 /// Updates the graph weight (EWMA) and last_scored after each.
 ///
 /// Returns the number of nodes scored and their (key, score) pairs.
-pub async fn score_memories_incremental(
+pub async fn score_memories_incremental<F, Fut>(
     context: &ContextState,
     max_age_secs: i64,
     response_window: usize,
     client: &ApiClient,
     agent: &std::sync::Arc<tokio::sync::Mutex<crate::agent::Agent>>,
-) -> anyhow::Result<Vec<(String, f64)>> {
+    mut on_score: F,
+) -> anyhow::Result<usize>
+where
+    F: FnMut(String, f64) -> Fut,
+    Fut: std::future::Future<Output = ()>,
+{
     let now = chrono::Utc::now().timestamp();
 
     // Collect unique memory keys with their first position
@@ -330,7 +335,7 @@ pub async fn score_memories_incremental(
     candidates.sort_by_key(|&(_, _, last)| last);
 
     let http = http_client();
-    let mut results = Vec::new();
+    let mut scored = 0;
 
     let total_entries = context.conversation.entries().len();
     let first_quarter = total_entries / 4;
@@ -355,7 +360,8 @@ pub async fn score_memories_incremental(
                 dbglog!(
                     "[scoring] {} max:{:.3} ({} responses)", key, max_div, n_responses,
                 );
-                results.push((key.clone(), max_div));
+                on_score(key.clone(), max_div).await;
+                scored += 1;
             }
             Err(e) => {
                 dbglog!(
@@ -365,7 +371,7 @@ pub async fn score_memories_incremental(
         }
     }
 
-    Ok(results)
+    Ok(scored)
 }
 
 // ── Fine-tuning scoring ─────────────────────────────────────────