Isn't AES encryption also reversible? The use-case for AES-SID appears to be mapping 64-bit ID values to 128-bit ciphertexts and reversibly producing an authenticated 64-bit ID value from the 128-bit ciphertext. Zero padded AES does both of these things as far as I can tell.

That by itself is an old problem; see for instance Black and Rogaway's paper on restricted domain ciphers:

https://web.cs.ucdavis.edu/~rogaway/papers/subset.pdf

It's also reversible:

newid = AES-EncryptBlock(0||id)

0||id = AES-DecryptBlock(newid)

Check for 8 zero bytes for authentication. Bonus: can use a constant instead of zeros for domain separation (eg different DB columns or tables).