[llvm] [memprof] Call llvm::SmallVector::reserve (PR #93324)

[Kazu Hirata via llvm-commits]

kazutakahirata wrote:

> lgtm. Any observable improvement to deserialization runtime?

Thank you for reviewing the patch!  Yes, improvements are measurable.  `perf stat -r5` tells me:

- -1.15% reduction on the number of cycles
- -0.28% reduction on the number of instructions



https://github.com/llvm/llvm-project/pull/93324