fix: observed-vs-expected share panel rendered empty

sum() strips labels while quantus_network_hashrate carries job and
instance, so the division matched nothing. Use scalar(), and take the
numerator from the hashes counter rather than the spiky gauge.

asset/grafana/quantus.json

@@ -9,7 +9,7 @@
   ],
   "timezone": "browser",
   "schemaVersion": 39,
-  "version": 8,
+  "version": 9,
   "refresh": "30s",
   "time": {
     "from": "now-6h",
@@ -1595,10 +1595,10 @@
             "uid": "prometheus"
           },
           "editorMode": "code",
-          "expr": "quantus_share_observed",
-          "legendFormat": "observed",
+          "refId": "A",
           "range": true,
-          "refId": "A"
+          "expr": "quantus_share_observed",
+          "legendFormat": "observed"
         },
         {
           "datasource": {
@@ -1606,10 +1606,10 @@
             "uid": "prometheus"
           },
           "editorMode": "code",
-          "expr": "sum(miner_hash_rate{job=\"quantus-miner\"}) / quantus_network_hashrate",
-          "legendFormat": "expected",
+          "refId": "B",
           "range": true,
-          "refId": "B"
+          "expr": "sum(rate(miner_hashes_total{job=\"quantus-miner\"}[5m])) / scalar(quantus_network_hashrate)",
+          "legendFormat": "expected"
         }
       ],
       "fieldConfig": {
@@ -1637,7 +1637,7 @@
           "sort": "desc"
         }
       },
-      "description": "THE panel. Expected is our hashrate over the network's; observed is the fraction of blocks we actually authored. Observed persistently below expected is orphan and stale-work loss \u2014 the only honest measure of whether connectivity tuning bought anything (issue #1). Needs hours of data to be more than noise, and both lines are meaningless until Syncing reads 0."
+      "description": "THE panel. Expected is our hashrate over the network's; observed is the fraction of blocks we actually authored. Observed persistently below expected is orphan and stale-work loss.\n\nscalar() is load-bearing: sum() strips labels while quantus_network_hashrate carries job/instance, so a plain division matches nothing and the panel renders empty \u2014 which is exactly how it shipped. And the numerator is the hashes counter, not the miner_hash_rate gauge, which spikes to impossible values after a restart.\n\nNeeds hours to mean anything, and both lines are meaningless while Syncing reads 1."
     },
     {
       "type": "timeseries",