Retry a failed 3D model inside the engine instead of in the layers

A mover set whose glTF or atlas fails to load retries by itself after
retryAfterMs (1,500 ms by default), at most three attempts, on Cesium,
Navara and the fake engine alike: failed() is true while the latest
attempt failed and stays true after the last, ready() turns true when a
retry loads, and an instance given another pickId loads again. Healthy
models of the same class no longer load again when one fails.

The layers drop what they did instead: moverBank's rebuilds, the loops
that dropped failed handles in flights, military and local ADS-B, and the
tracked-model backoff in the aircraft display. The contract suite checks
the one behaviour on both engines, with a mutant that never retries.

scripts/contract-mutants.test.mjs

@@ -207,7 +207,7 @@ const mutants = [
   ],
   [
     'failed model reports no failure',
-    'MoverSet.failed:',
+    'movers.createSet: failed model retries',
     (e) => {
       handle(e, 'movers', 'createSet', (set) => {
         set.failed = () => false;
@@ -216,13 +216,30 @@ const mutants = [
   ],
   [
     'failed model reports ready',
-    'MoverSet.failed:',
+    'movers.createSet: failed model retries',
     (e) => {
       handle(e, 'movers', 'createSet', (set) => {
         set.ready = () => true;
       });
     },
   ],
+  [
+    'failed model never retries',
+    'movers.createSet: failed model retries',
+    (e) => {
+      handle(e, 'movers', 'createSet', (set) => {
+        replace(set, 'commit', (commit) => () => {
+          const schedule = globalThis.setTimeout;
+          globalThis.setTimeout = () => null;
+          try {
+            commit();
+          } finally {
+            globalThis.setTimeout = schedule;
+          }
+        });
+      });
+    },
+  ],
   [
     'hidden mover reports shown',
     'movers.list: shown follows',