Cast: return a no-op ReconnectionService instead of null

[peterhel]

Cast: return a no-op ReconnectionService instead of null

Summary

CastDynamiteModuleImpl.newReconnectionServiceImpl currently returns null. But the
Cast SDK's ReconnectionService calls onCreate() on the returned delegate without a
null check, so any client that starts the ReconnectionService crashes with:

java.lang.RuntimeException: Unable to create service
  com.google.android.gms.cast.framework.ReconnectionService:
  java.lang.NullPointerException: Attempt to invoke interface method
  'void com.google.android.gms.cast.framework.zzq.onCreate()' on a null object reference
    at com.google.android.gms.cast.framework.ReconnectionService.onCreate(...play-services-cast-framework@@19.0.0:8)

The ReconnectionService is started on the connectionless (Cast.API_CXLESS) path that
modern first-party apps (Prime Video, YouTube, …) use, so in practice this crash-loops the
sender process during a cast session. The churn also surfaces downstream as a stream of
RemoteExceptions on the device-controller callbacks (onSendMessageSuccess /
onTextMessageReceived) — those are just callbacks landing on the listener of a process
that's being killed and restarted.

Fix

Return a minimal no-op IReconnectionService instead of null:

return new IReconnectionService.Stub() {
    @Override public void onCreate() {}
    @Override public int onStartCommand(Intent intent, int flags, int startId) {
        return android.app.Service.START_NOT_STICKY;
    }
    @Override public IBinder onBind(Intent intent) { return null; }
    @Override public void onDestroy() {}
};

microG doesn't implement Cast reconnection itself, so a no-op delegate is sufficient: it
satisfies the SDK's unconditional onCreate()/onStartCommand() calls without doing
anything, which keeps the session stable. (START_NOT_STICKY so the empty service isn't
re-created after being killed.)

Testing

On a Pixel 2 + a real Chromecast, casting a clip through the connectionless path:

• Before: the sender NPE-loops on ReconnectionService.onCreate; the device-controller
  callbacks throw RemoteException (dead listener); the session is unstable.
• After: no crash, no RemoteExceptions, the cast plays cleanly and the sender stays
  alive.

Notes

This is a microG-level gap independent of any single Cast PR — it affects every
connectionless Cast flow that wires up the ReconnectionService. Surfaced while
hardware-testing the open Cast PRs (#3377 in particular); with this fix in place those
connectionless implementations cast cleanly rather than crash-looping.