[libc++][pstl] Benchmark for parallel std::for_each (#222153)

Adds the first benchmark suite for PSTL algorithms - `std::for_each`.
It measures the algorithm against 3 different sizes: `64`, `64K` and
`64M` elements; and against 3 different per-element operation costs:
`minimal`, `cheap` and `expensive`, thus producing 9 separate
benchmarks.
The suite normally runs only with `std::execution::par`, serial
benchmarks can be additionally executed by defining
`TEST_PSTL_ENABLE_SEQ_BASELINES`.

GitOrigin-RevId: 1ac93e094347b45208f215c98038606432c87d60