[clang] [analyzer] Workaround for unintended slowdown (scope increase) (PR #136720)

[Donát Nagy via cfe-commits]

NagyDonat wrote:

@balazs-benics-sonarsource Unfortunately our current CI (which runs the static analyzer via CodeChecker) cannot provide file-level or entrypoint-level timing information, but I do have full analysis runtime measurements (which I'll edit into the table in my previous comment).

This comparison is between `legacy-inlining-prevention=false` (which is practically equivalent to the recent main revision a84a6f7dd68b218757e192fe21a806c80ef0b63d) and `legacy-inlining-prevention=true` (the default in this patch), but I can measure timing differences compared to other commits as well. I think the most natural comparison would be between
- this commit with its default `legacy-inlining-prevention=true` mode and
- reverting bb27d5e5c6b194a1440b8ac4e5ace68d0ee2a849 ("Don't assume third iteration") atop a84a6f7dd68b218757e192fe21a806c80ef0b63d (the parent of this commit)
because that way the comparison wouldn't be influenced by other commits that are merged in the last few months.

https://github.com/llvm/llvm-project/pull/136720