I see a lot of misconceptions here. But will be brief in reply.
Witness voting and "must-have-downvotes" are two entirely different processes. Yes, you can downvote any comment/post created by a witness. No, witness voting does not have a downvote mechanic. I think one needs not explaining why not in a "delegated proof of stake" system.
You underestimate the degree of automated spam abuse which has required so far resource credits and EIP already. Lowering dustvote rewards payout to 0.001 would be counter productive and only increase the earnings by such actors again.
A certain degree of abuse/maximization will always happen. It can not be fully nuked and that should not become the priority either. A system which is well-designed will minimize the scale of the abuse, with a scaling solution.
Emotional butthurt will always happen over downvotes. That proves a misunderstanding about the rewards belonging to the community until a post is payed out.
Humans bound to maximize will always find ways. Same with circle jerks. The latter are very hard to recognize as patterns without high levels of false positive and thus collateral automated damage. That is why also manual downvotes
Dapps, and future SMTs, can decide their own rewards system.
Sybill attacks (alt accounts) are a reality. If I can dedicate only 25% to an account, then I create 16 accounts if I want to maximize. (Don't get me started about IP recognition aso, maximizers know VPNs)
Also, there is no reward for activity. Unless you're an actifit user.
I understand that downvotes are a fickle thing to understand and accept. They are indeed weird and may seem hostile, poor design even, because entitlement over rewards can easily grow. But, the chain owes nobody anything until the final rewards amount is locked and subsequently paid out.
I too would much prefer that retaliation and similar sentiments were not part of the human modus operandi but they are. That still doesn't mean the system is flawed because many people fail to understand that nobody is owed anything by the chain. It's a weakness in the system because humans.
If you have a system which both understands the scale of (automated/spam) abuse and finding the middle path I will gladly read it.
So far you don't. That mostly because you don't seem to acknowledge the vast amount of maximization/abuse which needs countered as much as possible (because takes away potential rewards of real users).