reversing a hash is simply impossible and will always be impossible until the end of time
Yes. But there is another problem: it is always possible to find some preimage for any hash. RIPEMD-160 has 160 bits, secp256k1 has 256 bits, so assuming uniform distribution there are around 2^96 collisions for each address. And assuming that hash functions are regularly broken every few years, I really don't know what will be broken faster: hash functions used to generate addresses or elliptic curves.
(xk, yk) = k * (x1, y1)
Point multiplication in ECDSA is not that simple.