That's like saying locks are one more thing that can be broken, just like the doors. So why do people have locks on doors - in security you put up as many deterrents as practical and hope one of them saves the day. It's generally worth it.

That's actually the point of a sandbox. Code has bugs, so you sandbox it.

Now, the sandbox is also code, but to my understanding it's fewer LoC than what you are sandboxing. As bug rate per LoC is a fairly stable value, reducing LoC reduces total bug count. Ergo, by wrapping a large complex program with many LoC inside a small sandboxing function, you increase security (though it is not perfect, it still will have SOME bugs)

I don't think "security" is usually a binary concept, where it's either there or it isn't. It's more useful to draw an analog from the ratings of safes, which are along the lines of "it would take ___ hours / days to break this safe." Instead of lofty claims about being impervious to attack, put up obstacles and hurdles, that make attacks harder.