fix the remaining voice review gaps

keep analyst counts, speech, cards, cancellation, pointer capture, tool schema boundaries, and session lifetime consistent
add the exact Alps fallback and focused regression coverage

src/voice/gevRealtime.test.mjs

@@ -3733,6 +3733,16 @@ test('voice runs app actions itself and other tools through the catalog', async
     actions.push([name, args]);
     return { ok: true, action: name };
   };
+  const resets = [];
+  const disposals = [];
+  runner.resetConversation = (...args) => {
+    resets.push(args);
+    return 'reset';
+  };
+  runner.dispose = (...args) => {
+    disposals.push(args);
+    return 'disposed';
+  };
   const calls = [];
   let loads = 0;
   const catalog = {
@@ -3764,5 +3774,9 @@ test('voice runs app actions itself and other tools through the catalog', async
     actions.map(([name]) => name),
     ['zoom_to_globe', 'not_a_tool'],
   );
+  assert.equal(run.resetConversation('stopped'), 'reset');
+  assert.equal(run.dispose('removed'), 'disposed');
+  assert.deepEqual(resets, [['stopped']]);
+  assert.deepEqual(disposals, [['removed']]);
   assert.equal(withToolCatalog(runner, undefined), runner);
 });