[llvm] [ScheduleDAG] Fix and simplify the algorithm used for finding callseq_start (PR #149692)

[Min-Yih Hsu via llvm-commits]

mshockwave wrote:

> Are we emitting nested callseqs incorrectly then? The actual callseq emission code is not changed here

yeah let me see if I can emit a more sane callseqs instead

https://github.com/llvm/llvm-project/pull/149692