Only run 'adb root' when getting friendly name

* This avoid spamming adb root into the log often when we only need to run it
  once when a device is first seen.

renderdoc/android/android.cpp

@@ -584,12 +584,6 @@ extern "C" RENDERDOC_API void RENDERDOC_CC RENDERDOC_EnumerateAndroidDevices(rdc
       // Forward the ports so we can see if a remoteserver/captured app is already running.
       Android::adbForwardPorts(idx, tokens[0], 0, 0, true);
 
-      // run adb root now, so we hit any disconnection that we're going to before trying to connect.
-      // If we can't be root, this is cheap, if we're already root, this is cheap, if we can be root
-      // and this changes us it will block only the first time - and we expect this function to be
-      // slow-ish.
-      Android::adbExecCommand(tokens[0], "root");
-
       idx++;
     }
   }

renderdoc/android/android_utils.cpp

@@ -157,6 +157,15 @@ std::string GetFriendlyName(std::string deviceID)
   if(it != friendlyNameCache.end())
     return it->second;
 
+  // run adb root now, so we hit any disconnection that we're going to before trying to connect.
+  // If we can't be root, this is cheap, if we're already root, this is cheap, if we can be root
+  // and this changes us it will block only the first time - and we expect this function to be
+  // slow-ish.
+  //
+  // We do this here so that we sneakily take advantage of the above caching - otherwise we spam adb
+  // root commands into the log
+  Android::adbExecCommand(deviceID, "root");
+
   std::string manuf =
       trim(Android::adbExecCommand(deviceID, "shell getprop ro.product.manufacturer").strStdout);
   std::string model =