I believe a system should by secure by-design and not by anything else.
If that's possible I completely agree.
If it was the case that there is not possible secure modification of the PoW structure which is suitable to current ASICs, then maybe we better put with the weakness ...
Unfortunately suitable course of action isn't necessarily a black and white call. If we were in a lab under controlled conditions then of course we could use pure logic for system optimization. This is not the case. Bitcoin is made up of independent actors all of which to some degree influence the whole. It's similar to how economics as a science is the one field where experts can agree upon a set of data yet come up with opposing views.
With Bitcoin protocol changes therefore can't ever be taken lightly. The larger the system grows the more difficult it is to come to consensus on any change. At some point I imagine it becomes impossible (one reason I believe alt-coins have a role). So I'm very glad people like you are raising these ideas sooner than later.
Bitcoin as money relies on shared confidence in its value. Confidence in Bitcoin's value naturally rests upon many things including the viability of the system, which includes withstanding attacks. It's unclear what would be the result of a pushed for yet unsuccessful protocol change. That's why we need to be careful about what things are viewed as critical.
(there are other ways to counter block discarding attacks, such as my "fork punishment" proposal).
While I agree with many of your conclusions I'm not yet sold on the fork punishment countermeasure.
I have no intention to argue about off-topic issues that have been thoroughly discussed in other threads, yet I will just note that there is an incentive for selfish mining: increased mining profits. Such an attack will most probably not cause the total destruction of the system ...
Even you yourself say "probably" not cause the total destruction of the system. Nobody
can say what the effect would be. Something less than completely healthy seems assured, and that in all likelihood means lower value for coins.
Moreover, I dont think we should make any assumptions about the attacker intentions and incentives: the attacker may theoretically want to harm the system (say if it is a terror organization, the Chinese government, or more likely the NSA).
But that matters in practice. The security of the system relies on the assumption an attacker's resources will be low compared to the honest network. There are real costs (including other than financial) to prepare and execute the attack. If an attacker can't use pooled resources from the network their challenge is great. Governments might be the only viable candidate but governments are subject to political repercussion. Any attacker building out the server farms, power centers, etc. would probably not go undetected.