[clang] [Serialization] Load Specializations Lazily (PR #76774)

[Chuanqi Xu via cfe-commits]

ChuanqiXu9 wrote:

> Have you run that patch on bigger workflows? Do we have some performance numbers to compare?

I've tested it functionality in our largest workload about modules. It runs well. But our uses of modules don't have a lot of complexities while it has a large scale. For performances, I plan to make it this week. It is a little bit additional work since I need to compile the compiler with different optimizations to have a fair comparison.

https://github.com/llvm/llvm-project/pull/76774