Long post is long.
I like most of these ideas, but my main worry is some of the ideas is moving from "emitters specifying global particle behavior", to "directly specifying individual particle behavior".
I.e., after being able to: control random subsets of particles, control specific subsets of particles, and access specific particle data, we are very close to directly manipulating particles. The main reason the particle system 'works' is because it avoids doing that and can perform reasonably well as a result.
The rest looks good imo (though I don't use terrain stuff so I can't really comment there).