Fix SM4-CBC regression on Armv8

Fix SM4-CBC regression on Armv8

Fixes #19858

During decryption, the last ciphertext is not fed to next block
correctly when the number of input blocks is exactly 4. Fix this
and add the corresponding test cases.

Thanks xu-yi-zhou for reporting this issue and proposing the fix.

Reviewed-by: Tomas Mraz <tomas@openssl.org>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/19872)