Reduce context note preview sizes to cut prefill latency

Auto-notes (keyword-matched): 2000 → 800 chars each (×3 max = 6000 → 2400 chars).
Pinned note (explicit context): was unbounded → capped at 4000 chars with [truncated] marker.

The main post-GPU bottleneck is TTFT caused by the prefill phase — the model
processing the full input before generating any tokens. Shorter context =
faster prefill. Users can ask follow-up questions for more detail.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

src/fabledassistant/services/llm.py

@@ -289,16 +289,20 @@ async def build_context(
         "auto_notes": [],
     }
 
-    # Include current note context if provided
+    # Include current note context if provided (cap at 4000 chars — explicitly selected
+    # so more generous, but very long notes would blow up the prefill phase)
     if current_note_id:
         note = await get_note(user_id, current_note_id)
         if note:
             context_meta["context_note_id"] = note.id
             context_meta["context_note_title"] = note.title
+            body = note.body or ""
+            truncated = body[:4000]
+            suffix = "\n[...truncated]" if len(body) > 4000 else ""
             system_parts.append(
                 f"\n\n--- Current Note ---\n"
                 f"Title: {note.title}\n"
-                f"Content:\n{note.body}\n"
+                f"Content:\n{truncated}{suffix}\n"
                 f"--- End Note ---"
             )
 
@@ -314,7 +318,9 @@ async def build_context(
             )
             snippets: list[str] = []
             for n in notes:
-                body_preview = n.body[:2000] if n.body else ""
+                # 800 chars keeps context meaningful without bloating the prefill phase.
+                # The user can ask follow-up questions for more detail.
+                body_preview = n.body[:800] if n.body else ""
                 snippets.append(f"- {n.title}: {body_preview}")
                 context_meta["auto_notes"].append({"id": n.id, "title": n.title})
             if snippets: