[clang-tidy] Prefer the faster LLVM ADT sets and maps over `std::` ones (#174357)

The LLVM docs give a good description of [why `std::` containers are
slower than LLVM
alternatives](https://llvm.org/docs/ProgrammersManual.html#set). To see
what difference switching to the LLVM ones made, I [reused the
approach](https://github.com/llvm/llvm-project/pull/174237#issuecomment-3707395449)
of measuring how long it takes to run all checks over all standard
library headers (MSVC STL in my case). Using hyperfine (which basically
runs a program multiple times and computes how long it took):
```sh
hyperfine --shell=none './build/release/bin/clang-tidy --checks=* all_headers.cpp -header-filter=.* -system-headers -- -std=c++23'
```
...the results were:

Before:
```
Benchmark 1: ./build/release/bin/clang-tidy --checks=* all_headers.cpp -header-filter=.* -system-headers -- -std=c++23
  Time (mean ± σ):     53.253 s ±  0.089 s    [User: 46.480 s, System: 6.748 s]
  Range (min … max):   53.118 s … 53.440 s    10 runs
```
After:
```txt
Benchmark 1: ./build/release/bin/clang-tidy --checks=* all_headers.cpp -header-filter=.* -system-headers -- -std=c++23
  Time (mean ± σ):     51.798 s ±  0.126 s    [User: 45.194 s, System: 6.575 s]
  Range (min … max):   51.620 s … 51.995 s    10 runs
```

...which is a nice little speedup for just switching some containers. I
didn't investigate which checks in particular were the source of the
speedup though.