[llvm] [IR] Use block numbers in PredIteratorCache (PR #101885)

[Alexis Engelke via llvm-commits]

aengelke wrote:

(Meh, didn't intend to push my experiment to this branch... just wanted to get performance results (perf/...) for a two-level scheme with a BitVector to reduce memset overhead.)

https://github.com/llvm/llvm-project/pull/101885