[NewOffloadModel] Add support for passing backend options for multiple device architectures

[YixingZhang007]

This patch adds the support of passing different backend options through -Xsycl-target-backend for each of multiple device architectures.
An example is like the following, where we are passing different options for pvc and dg1.

clang++ ... -fsycl-targets=intel_gpu_dg1,spir64_gen \
  -Xsycl-target-backend=spir64_gen "-device pvc -options -extraopt_pvc" \
  -Xsycl-target-backend=intel_gpu_dg1 "-options -extraopt_dg1" ...

The following changes are made to support this feature:

1. In SYCL.cpp, add a new function extractDeviceFromArg for extracting the device from the backend option when the SYCL target is spir64_gen. This allows finding the device architecture from arguments such as: -Xsycl-target-backend=spir64_gen "-device pvc -options -extraopt_pvc"
2. In Clang.cpp, we construct a --device-compiler=... argument to pass into clang-linker-wrapper for each SYCL target. The --device-compiler contains a -device <arch> to differentiate the device that the backend option is specified for. When the SYCL target is spir64_gen (not intel_gpu_arch), we call extractDeviceFromArg to extract the device architecture.
3. clang-linker-wrapper is updated to filter OPT_device_compiler_args_EQ based on the architecture, and specify each architecture when calling sycl-post-link.
4. sycl-post-link is updated to support multiple architectures being passed in, such that cases like: -Xsycl-target-backend=spir64_gen "-device arch1,arch2"
5. Added a new test in sycl-offload-new-driver.cpp to verify the changes.