[llvm] [ci] limit parallel windows compile jobs to 24 (PR #93329)

[Lucile Rose Nihlen via llvm-commits]

lnihlen wrote:

If compiling flang really does require >8 GB per .cpp file on MSVC, it probably makes sense to drop this to 16 threads. Although none of the analytics show memory pressure as a problem (see the chart I posted in the Discourse thread), I'm wondering if we're running slowly because we're swap thrashing, because reducing paralellism doesn't seem to negatively impact build times, at least based on my limited testing.

https://github.com/llvm/llvm-project/pull/93329