[llvm] [DFAJumpThreading] Enable DFAJumpThread by default. (PR #157646)

[Hongyu Chen via llvm-commits]

XChy wrote:

I believe https://github.com/llvm/llvm-project/pull/161632 has resolved most outliers. The bottleneck mostly lies in the later optimizations in the pipeline after duplicating blocks.
Another point to improve is the preservation of BranchFrequencyInfo.

https://github.com/llvm/llvm-project/pull/157646