Users consent to the website seeing their traffic and the website consents to Cloudflare doing the SSL termination. This isn't too much different from the website consenting to analytics scripts monitoring webpage activity (i.e. Hotjar). If they did something shady, then users & the website would both be rightfully mad at them. But Cloudflare hasn't, so far at least.
Meanwhile, Facebook is known to do literally everything shady that is possible to do with a user's data, as well as plenty of things that weren't even a thing before they invented entirely new methods of tracking and selling data, so it's rightfully insane to trust them with anything, especially website traffic that they have no rights to.
Only if you let them manage the SSL connection. Load balancers can easily relay individual TCP connections that are encrypted - load balancing doesn't require decryption.
> And even if not, snooping on VMs is pretty trivial.
They'd have to go out of their way to do this, and this would probably be the end of them if it were ever found out. So it's safe to assume any provider who wants to continue existing will not be doing this.
I did work for a public cloud and we did think of VMI for diagnostics and malware checks. Once deployed and automated, it would be trivial to reuse for other purposes. I don't expect public cloud to use that daily, but I'd be surprised if they didn't have the process ready.
On the other hand, you want to process the LB traffic as fast as you can and any monitoring/reporting delay would have bad effects. Reconfiguring the filters / sinks at runtime takes effort too.
With experience in both areas, I can tell you they're comparable overall. You have to go out of your way to do it, but it's not too far.
Users consent to Facebook seeing their traffic and it's suddenly a problem?