I think requirement 2 is a problem here.
If your cloud provider has the plaintext how can you trust it?
If your cloud provider has the plaintext how can you trust it?
For example, suppose that this was implemented on a VPS on some random hosting provider, and I own all of the code on the VPS. The hosting provider could record all email traffic coming into that box, but that problem isn't in scope (and could be addressed with, say, openPGP).
Who/what are you trying to defend against?
Basically, if someone takes control of the hardware or software we're hoping that it would still be very difficult to get at the contents of the emails.