[clang] [llvm] [OpenMP] Remove 'libomptarget.devicertl.a' fatbinary and use static library (PR #126143)

[Michał Górny via cfe-commits]

mgorny wrote:

And after switching to clang, it fails anyway:

```
$ /usr/lib/ccache/bin/x86_64-pc-linux-gnu-clang++ -O2 -pipe -march=native -Wl,-O1 -Wl,--as-needed -Wl,-z,pack-relative-relocs    --target=nvptx64-nvidia-cuda -r -nostdlib -flto -Wl,--lto-emit-llvm DeviceRTL/Allocator.cpp-nvptx.o DeviceRTL/Configuration.cpp-nvptx.o DeviceRTL/Debug.cpp-nvptx.o DeviceRTL/Kernel.cpp-nvptx.o DeviceRTL/LibC.cpp-nvptx.o DeviceRTL/Mapping.cpp-nvptx.o DeviceRTL/Misc.cpp-nvptx.o DeviceRTL/Parallelism.cpp-nvptx.o DeviceRTL/Profiling.cpp-nvptx.o DeviceRTL/Reduction.cpp-nvptx.o DeviceRTL/State.cpp-nvptx.o DeviceRTL/Synchronization.cpp-nvptx.o DeviceRTL/Tasking.cpp-nvptx.o DeviceRTL/DeviceUtils.cpp-nvptx.o DeviceRTL/Workshare.cpp-nvptx.o -o libomptarget-nvptx.bc
x86_64-pc-linux-gnu-clang++: error: cannot determine nvptx64 architecture: /usr/lib/llvm/20/bin/nvptx-arch: ; consider passing it via '-march'; environment variable CLANG_TOOLCHAIN_PROGRAM_TIMEOUT specifies the tool timeout (integer secs, <=0 is infinite)
```

(I have no nVidia GPU but I don't think host GPU should be affecting offload build)

https://github.com/llvm/llvm-project/pull/126143