You're wrong because it's a strawman since if an attacker can intercept the hash they could as easily intercept the plaintext in your traditional server-side architecture. The attacker cannot replay a TLS unless there is a problem with it, there have been many issues in TLS stacks, but it's the most widely deployed.

Furthermore, using plaintext any further from the owner or exposed longer than is necessary is inherently less secure because your breach of https would also compromise users' passwords.