Warm welcome to the community and great post!
In answer to your first question, you can absolutely make attachments between objects that are intersecting each other at rest. You'll most likely have to turn collisions off on the bones if you want to keep the muscle-muscle collisions. But you could for example have sliding attachments to the bones that you're not colliding with if you choose that approach.
You might actually find that switching collisions back on for the bones won't be as destructive as you're potentially expecting. So long as none of the points of mesh A are further than half way through the volume of mesh B (and vice versa) then Ziva will probably resolve the intersection on the first frame anyway. This will however change the silhouette of your model as the intersecting muscles push outward initially.
It might be worth getting the eval once you feel like you have the broad strokes of the model done, so you can run some tests to see what you can get away with.
Your assumption about the fat pass is correct. You solve the fat as a single Ziva tissue that rides on top of the fascia.
You might be able to get some insights about fat distribution from cross-sectional MRIs.. This sort of thing:
Or images like this:
Looking forward to seeing your progress and hearing from you soon.