[llvm] [BOLT] Optimize basic block loops to avoid n^2 loop (PR #156243)

[Jakub Beránek via llvm-commits]

Kobzol wrote:

The size of `.text.warm` back then was `841 250` B (~840 KiB). The absolute max RSS wins are shown [here](https://perf.rust-lang.org/compare.html?start=774ae599aba35f7b323e77be88923b8f476ee783&end=9023f908cfbe7a475f369717a61cb8eb865cfd25&stat=max-rss&showRawData=true). They are in the range of ~2 MiB to 30 MiB.

https://github.com/llvm/llvm-project/pull/156243