The randomness is not enough for IND-CCA-2. If we get a message (so $L$ bits of data plus the $y$ to reconstruct the random seed), we can modify it, say flip the first bit, and ask the decryption oracle for a decryption of the modified message (which will have the same $y$!), which we will get. Then the original message can easily be obtained: we get the same key stream for the message (because we have the same $y$ in the modified message), we xor the plain text of the modified message with the modified message and we get the key stream for both the original challenge and (by construction) the modified message. So now we can decrypt the original message too.
Or in a more usual game: we choose a message $p$ to be encrypted and we get back either random or the actual encryption. We do out procedure from above and see if what we got back decrypts to our $p$, and if not, we "expose" the player as having "played" random.
In short, we can ask for a modified message decryption and get useful info. This is not allowed to happen in a IND-CCA-2 secure system.