I am a supporter of Autocrypt. As Monperrus correctly wrote in the OP, Autocrypt is mainly for "disseminating end-to-end encryption in the email realm".
In the Debian bug You linked, You wrote "sending sensitive information unencrypted while expected it being encrypted might be dangerous.".
Securing sensitive information is not the Autocrypt approach. Autocrypt follows the new perspective of opportunistic security: Encrypt if possible, otherwise send unencrypted, but make it easy for the mass of users to get encryption disseminated in a first step.
Nevertheless, I agree that if Autocrypt in TB 78 does less encryption than with Enigmail, this should be improved. Clear text is not wrong, but more encryption is better ;)