Fix: Consistent Usage of target_link_libraries for CUDA Targets

[basnijholt]

This pull request addresses a build configuration issue that arises (introduced in #643) when mixing plain and keyword signatures of target_link_libraries in CMake for the qsim_cuda, qsim_custatevec, qsim_decide targets. The existing setup led to a CMake error, as documented:

CMake Error at pybind_interface/cuda/CMakeLists.txt:48 (target_link_libraries):
  The plain signature for target_link_libraries has already been used with
  the target "qsim_cuda".  All uses of target_link_libraries with a target
  must be either all-keyword or all-plain.

Changes Made:

• Modified the target_link_libraries command for the qsim_cuda, qsim_custatevec, qsim_decide targets to use the plain form. This change aligns with the internal usage of target_link_libraries within the cuda_add_library macro, ensuring consistency across the build configuration.

Rationale:

CMake requires that all invocations of target_link_libraries for a specific target maintain a consistent usage pattern - either all using the plain form or all using the keyword form (such as PUBLIC or PRIVATE). The cuda_add_library macro used in this project employs the plain form of target_link_libraries, which necessitates a similar approach for additional invocations of target_link_libraries for the qsim_cuda, qsim_custatevec, qsim_decide targets.

By aligning the usage pattern, this pull request resolves the build configuration error and maintains consistency in the CMake setup, ensuring a smoother build process.

[basnijholt]

@NoureldinYosri note that the CUDA build is currently broken and these changes are required to make it work.