cmd/internal/obj: implement auto-nosplit for arm64, ppc64

[rsc]

See #11482 and CL 17165. The fix was to add nosplit tags, but the implication is that arm64 and ppc64 do not have the same "auto-nosplit" optimization that the other architectures do for leaf functions with tiny frames. They should.

[mwhudson]

I think it's only cmd/internal/obj/x86 that implements this optimization, i.e. arm and mips need this too.

[gopherbot]

CL https://golang.org/cl/31357 mentions this issue.

[randall77]

Also 386. (amd64 has this optimization, but 386 doesn't.)

[gopherbot]

Change https://golang.org/cl/302853 mentions this issue: cmd/internal/obj/arm64: mark functions with small stacks NOSPLIT