Form Herds from Multiple scf.forall nests.

[SamuelBayliss]

In order to support code generated from loop-peeling, we'd like to look at forming herds from multiple scf.forall statements ( rather than targeting just one), so that L1 allocations can stay local within a herd definition.

[yzhang93]

Here's a proposal nod-ai/iree-amd-aie#210

[yzhang93]

I have an example showing IR before peeling, after peeling and after mapping to a herd with core ids https://gist.github.com/yzhang93/7ab35c4da0f958d4ffaca31977b91c7f. We can have a discussion based on this.