[llvm] [CI] Disable flang from pre-commit tests when Flang files are not touched (PR #93485)

[Renato Golin via llvm-commits]

rengolin wrote:

> This is a bit of a weird take to me: what does "officially in production" mean? Did MLIR ever got "officially in production" in the context of the LLVM project? The only thing I'm aware of is the notion of "experimental", which applies to backends (Flang was kind of in-between at the beginning because it was C++17 before LLVM was, but it's all aligned now). There are degrees of maturity of course, but I haven't seen issues with the Flang-on-Linux CI so far (some frontend files are annoyingly heavy to compile!), or only very occasionally (flaky tests can happen, but not more than in LLVM itself or any other project)

Our main support policy is: https://llvm.org/docs/SupportPolicy.html I have reported at least three serious issues with Flang testing since last year, all of which took over a month to get fixed. This one is only the last example.

**But, this PR is NOT about that at all. This is a tangent that we should not pursue. The core issue here is PRE-COMMIT versus POST-COMMIT testing.**

I think we need to be serious about pre-commit CI. This issue was causing grief for a month to a lot of people who do not care about Flang or cannot fix the underlying issue. It will not make it less tested if it has post-commit testing, it will just have a slightly longer cycle to fix rarer problems.

https://github.com/llvm/llvm-project/pull/93485