[llvm] [llvm] Lower latency bonus threshold in function specialization. (PR #143954)

[Slava Zakharin via llvm-commits]

vzakhari wrote:

@kiranchandramohan , @tblah this change does not affect LTO performance on aarch64 in my testing, but please let me know if you see any discrepancies.  Also feel free to revert it while I am out of the office from 19th to 23rd of June.

https://github.com/llvm/llvm-project/pull/143954