So, I'm not really a fan of "omnibus bills" because it becomes hard to isolate which part of the changes were responsible for which outcomes. Isolated variables and all that. Yet I do understand trying to avoid too many hardforks.
So in general I say YES, with the following caveats.
For 1, I'm fairly agnostic and don't have too strong reasons to say yes or no with confidence. I'd also be interested in a future change allowing an author to choose any % higher than the minimum should they desire -- for example, instead of declining rewards I could opt to give everything to curators. To avoid future hardforks, the implementation should be a witness parameter to define the minimum an author can use.
For 2, choosing a good constant will be important, and perhaps should be defined in terms of the general economy or be a witness parameter. However, given a solid design here, I am in support for using a reward curve like this.
A long time ago I suggested using a relu / rectifier like soft plus on rewards. In machine learning, rectifiers are used to remove noise from a system by damping lower values, while leaving larger ones alone. The proposed change has a similar effect.
For 3, I have been a big supporter of a downvote pool for a long time. I would much prefer the availability % to be a witness parameter rather than fixed, however, as I don't agree that 25% is the right value (I believe it should be 100).
One fairly important part here however is separate delegation of upvote power and downvote power.
As an example, I may wish to delegate a portion of my upvote power to a curator looking to support Japanese authors, and delegate a portion of my downvote power to a collective that indiscriminately downvotes to counter bidbot upvotes. Like in this example, it doesn't make sense to restrict the two choices into the same entity.
In this case, there is more to combining the items than just reducing hardforks. They directly interact. For example, reducing the cost of downvotes makes it more likely those seeking to avoid downvotes will shift to harder-to-find micropayouts. Thus #3 is the primary motivation for #2. Likewise, the game theoretic balance between curation earnings and self-voting earnings depends on the relative risk of being downvoted, so the size of the downvote pool (#3) and the curation share (#1) are not independent variables.
IMO one can reasonably evaluate this entire proposal as a coherent package and either accept it or reject it without considering it as an onmibus package of otherwise-independent items (as we have been asked to do in the past).