[llvm] [flang-rt] Optimise ShallowCopy and use it in CopyInAssign (PR #140569)

[Michał Górny via llvm-commits]

mgorny wrote:

Not really — I've just tried building the latest versions (as of 8d374f1f5bd18e23f94e377e544a1b04de993973). The most recent snapshot (`21.0.0_pre20250523`) should also be affected, but it's a bit out of date due to other regressions. My first guess is that it happens only with GCC, but I haven't tried Clang yet.

https://github.com/llvm/llvm-project/pull/140569