Neither RIPEMD-160 nor SHA256 are subject to such attack. They are not analytical and only a brute force attack is feasible to be run by adversaries which is not practical and will not be practical in foreseeable future, hence, they are safe now.
Yes SHA256 and RIPEMD160 algorithms are safe today but even if they are not linked to large number arithmetic, there is not proof that they cannot be reversed or predicted in polynomial time and space. As for ECDSA, they is no proof that ECDLP cannot be solved. Today the security of ECDLP256 is ~128bit and 160bit for RIPEMD160. Both are not feasible today but the probability that someone find a way to solve ECDLP256 or to reverse hashing algorithms is not zero. It is not possible to predict which algorithm will be defeated first.
There is no objective reason to say that exposing ECDSA public key for a long time is less safe than exposing an address.
I'm not comfortable with this argument. ECDLP has been discredited by Shor's algorithm which offers polynomial time/space solution for a QC based machine, the very nature of discrete logarithm problem is fragile and vulnerable to further mathematical developments just like what happened with Shor algorithm and QC vulnerability, it is not exactly the case for SHA256 or RIPEMD160 we have no reason to be worried about them to break and if anybody has any concern about such a possibility even in next couple of centuries s/he should stop using bitcoin as a store of value.