[compiler-rt] [scudo] Split the code path of memory tagging out from allocate() and quarantineOrDeallocateChunk() (PR #83493)

[via llvm-commits]

ChiaHungDuan wrote:

I measured the cache miss rate by using simpleperf and running [malloc-rss-benchmark](https://android.googlesource.com/platform//bionic/+/ee824ae6e01a16565ae8b7dd078ee51530429c68/benchmarks/malloc_rss_benchmark.cpp) (Remove some noises which may impact the result) for 20 times each.

I don't see significant difference (0.5934682 vs 0.5927589, less than 0.5%). As mentioned in the [comment](https://reviews.llvm.org/D159392#4651499), this change helps with readability in both source and asm levels. As a result, I think it's still a good reason to do the clean up.

BTW, there were two changes and I put them in the same pull request.

https://github.com/llvm/llvm-project/pull/83493