Add support for different GPU operations (MPI, CCL)

**Is your feature request related to a problem? Please describe.**

**Describe the solution you'd like**
Currently this version use oneCCL as the backend for Allreduce operations, and I see the `HOROVOD_GPU_OPERATIONS ` is set to `CCL` in `CMakeLists.txt`.
```
if(CMAKE_CXX_COMPILER MATCHES "icpx$")
set(ENV{HOROVOD_GPU} "SYCL")
set(ENV{HOROVOD_GPU_OPERATIONS} "CCL")
set(ENV{HOROVOD_WITHOUT_GLOO} "1")
endif()
```

However, I want to use the pure Intel MPI (MPI_Allreduce) as the backend but it doesn't work even I change it to `MPI`.
After compiling, the `hoeovodrun --check-build` just shows the following. It seems like the backends are disables.
```
Horovod v0.28.1.4:

Available Frameworks:
[ ] TensorFlow
[X] PyTorch
[ ] MXNet

Available Controllers:
[ ] MPI
[ ] Gloo

Available Tensor Operations:
[ ] NCCL
[ ] DDL
[ ] CCL
[ ] MPI
[ ] Gloo
```
How can I run Horovod with MPI instead of oneCCL?

**Describe alternatives you've considered**
Add support for different GPU operations (MPI, CCL)

**Additional context**