fix(combo): keep nested combos as single units and stop 0-token detail rows

Nested combos (comboA lists comboB, comboC, …) now stay one slot each:
the inner combo always runs as fallback to produce a single answer.
Failed hops are no longer written to Details/usage, and streaming no
longer inserts a 0-token placeholder row.

- chat.js: comboStack cycle detection; nested combos forced to fallback;
  persistUsage="success-only" for combo hops
- combo.js: discardResponse() cancels unused bodies (fusion timeout /
  fallback) so dropped streams fire onStreamComplete; getComboModelsFromData
  keeps nested names and honors enabled=false
- requestDetail.js: tokensForDetail() canonicalizes Claude/Gemini usage;
  shouldPersistRequestDetail() skips streaming-start and non-success hops
- streamingHandler.js: drop the 0-token streaming placeholder write
- RequestDetailsTab.js: read Gemini/Claude token names; show "streaming"
  status in amber
- tests: add combo-nested.test.js (13 cases)
- gitignore: ignore local .vitest/ artifacts

Co-authored-by: CommandCodeBot <noreply@commandcode.ai>

open-sse/services/combo.js

@@ -246,23 +246,39 @@ export function resetComboRotation(comboName) {
 }
 
 /**
- * Get combo models from combos data
+ * Cancel an unused Response body so its SSE transform (and usage callback)
+ * is not left hanging. Fusion timeouts and combo fallback both drop Response
+ * objects without reading them; an unread stream never fires onStreamComplete,
+ * which is how request-details rows get stuck at input/output tokens = 0.
+ */
+export function discardResponse(res) {
+  if (!res || res.__timeout || res.__error) return;
+  const cancel = res.body?.cancel;
+  if (typeof cancel === "function") {
+    try { cancel.call(res.body); } catch { /* already closed / consumed */ }
+  }
+}
+
+/**
+ * Get combo models from combos data.
+ * Nested combo names are kept as-is (comboA listing comboB yields ["comboB", ...]).
+ * Flattening them into leaves would explode fusion/round-robin: every nested
+ * leaf becomes its own panel/rotation slot. Nested combos run as one unit
+ * (see handleSingleModelChat: inner strategy forced to fallback).
+ *
  * @param {string} modelStr - Model string to check
  * @param {Array|Object} combosData - Array of combos or object with combos
- * @returns {string[]|null} Array of models or null if not a combo
+ * @returns {string[]|null} Direct members, or null if not a combo
  */
 export function getComboModelsFromData(modelStr, combosData) {
-  // Don't check if it's in provider/model format
-  if (modelStr.includes("/")) return null;
-  
-  // Handle both array and object formats
+  if (typeof modelStr !== "string" || modelStr.includes("/")) return null;
+
   const combos = Array.isArray(combosData) ? combosData : (combosData?.combos || []);
-  
-  const combo = combos.find(c => c.name === modelStr);
-  if (combo && combo.models && combo.models.length > 0) {
-    return combo.models;
+  const combo = combos.find((c) => c.name === modelStr);
+  if (!combo || combo.enabled === false || !Array.isArray(combo.models) || combo.models.length === 0) {
+    return null;
   }
-  return null;
+  return combo.models;
 }
 
 /**
@@ -339,6 +355,9 @@ export async function handleComboChat({ body, models, handleSingleModel, log, co
         return result;
       }
 
+      // Falling through: this body will never be read by the client.
+      discardResponse(result);
+
       // For transient errors (503/502/504), wait for cooldown before falling through
       // so a briefly-overloaded provider gets a chance to recover rather than being
       // skipped immediately (fixes: combo falls through on transient 503)
@@ -509,8 +528,13 @@ function collectPanel(calls, { minPanel, stragglerGraceMs, panelHardTimeoutMs })
     const hardTimer = setTimeout(finish, panelHardTimeoutMs);
     calls.forEach((p, i) => {
       Promise.resolve(p)
-        .then((v) => { out[i] = v; })
-        .catch((e) => { out[i] = { __error: e }; })
+        .then((v) => {
+          // Quorum/timeout already moved on: cancel the late body so its
+          // streaming usage callback is not left at tokens=0 forever.
+          if (finished) discardResponse(v);
+          else out[i] = v;
+        })
+        .catch((e) => { if (!finished) out[i] = { __error: e }; })
         .finally(() => {
           settled++;
           if (out[i] && out[i].ok) ok++;
@@ -590,7 +614,7 @@ export async function handleFusionChat({ body, models, handleSingleModel, log, c
     if (!res) { log.warn("FUSION", `Panel ${model} dropped (straggler/timeout)`); continue; }
     if (res.__timeout) { log.warn("FUSION", `Panel ${model} timed out`); continue; }
     if (res.__error) { log.warn("FUSION", `Panel ${model} threw`, { error: res.__error?.message || String(res.__error) }); continue; }
-    if (!res.ok) { log.warn("FUSION", `Panel ${model} failed`, { status: res.status }); continue; }
+    if (!res.ok) { log.warn("FUSION", `Panel ${model} failed`, { status: res.status }); discardResponse(res); continue; }
     try {
       const json = await res.clone().json();
       const text = extractPanelText(json);
@@ -602,6 +626,11 @@ export async function handleFusionChat({ body, models, handleSingleModel, log, c
       }
     } catch (e) {
       log.warn("FUSION", `Panel ${model} unparseable`, { error: e.message || String(e) });
+    } finally {
+      // Panel responses are never returned to the client; cancel the original
+      // body (clone().json() already captured the payload) so streaming
+      // placeholders are not left as success/0-token rows.
+      discardResponse(res);
     }
   }