I think your points 1,2, and 3 are the most critical. Along with #3, I want to decrease the time of the voting window.
How long does it take a piece of content to propagate to 90+% of its reach? Less than two days. People share content very quickly, and there are some adjustments for different time zones in a global market. If you look at any other content platform you won't find old (2+ days) content above the fold.
So, change the voting window to 2 days and that allows your power down time to be shortened to 2 days + some number of blocks.
In 2016/17, there were two payout periods, short and long. The long period was 30 days, and the short one was either 12 or 24 hours. In the end, they moved to one period which is a compromise of both.
It was both 12 and 24 at different times. 24 worked better because of time zones.
I don't think 7 days has worked out in a useful way (the later portion of the 7 days has mostly just become a way to hide milking rewards, as was the 30 days when that existed) and would be supportive of going back to 24 hours. However, the vote recharge and mana stuff also needs some lockup period to avoid similar issues of vote then move stake and vote again.
I had discussed this with curators a while ago. Most agreed that 48 hours is enough time for them to find and curate posts, so I'd be in favour of that. 24 hours is better than 7 days too, though some "sleeper hits" content will be missed. Ultimately, I'd rather the focus be on removing the common reward pool entirely, so we don't need to waste time on these. SMTs can experiment with their own payout times if so inclined.
Unfortunately it has become a common and lazy habit to claim that SMTs are going to fix everything. In fact, a lot about SMTs is fixed using the same mechanism as the global pool, and not arbitrarily configurable, at least in its initial implementation, which doesn't even exist yet, and god knows how long it will take to not only be completed but to reach a level of quality and reliability that can be responsibly deployed. I don't recall if payout time was one of them but I think it very well might be.
A more fully configurable vision of SMTs where each and every little aspect of behavior can be configured on a per-SMT basis is a pipe dream that could be years or even decades away given the philosophy behind SMTs. A far better approach would be abandon this idea of hard coding everything into the blockchain itself and go back to the original plan for smart contracts where different projects can build their own SMT (or whatever you want to call it) to work however they like.
Some sleeper hits are always missed by the way. I can think of one case where an interesting post was pointed out to me months later and received near-zero votes originally. Any payout time is going to be a compromise, particularly when power down time is tied to it (though until we convince enough people that power down time needs to be short, this doesn't matter in practice).
Let me be very clear - I have very low expectations from SMTs, particularly the spec proposed by Steemit. I haven't followed development so I don't know how it has evolved. The Steem Engine and tribes stuff is a tangential prototype (and is arguably more sophisticated already) that suggests what we already know - it's very, very unlikely it'll have any significant positive impact. The social media space is mature and saturated, even if one had deep pockets and sought to revolutionize the market, there's basically no rational reason for building it atop Steem.
I misused "SMT" to mean the current spec, as you correctly point out. To clarify what I said above, I want admins and moderators to have the tools required to configure their own communities and tokenisation however they want. I'm well aware this is not what "SMT" means currently and will take years to develop, as I made clear in the OP. I absolutely agree these should be fluid smart contracts rather than hardcoding everything to the blockchain, which has no practical value beyond some misguided idealism.
None of this will make Steem a "Facebook killer" - that's simply delusional. But we are at a dead end with the current paradigm, in my opinion, and must try something different.