Enhance OpenACC Support in GCC

GCC currently supports most OpenACC directives for GPU offloading, but several features remain unimplemented or incomplete. This project targets four concrete gaps in the OpenACC implementation within GCC: - init, shutdown, and set directives (OpenACC 2.5+): Implementing the directive-form equivalents of the already-available runtime API routines, following the pattern of the existing #pragma acc wait infrastructure. - bind clause: Extending full support to C, C++, and Fortran by mapping the clause to OpenMP declare target and declare variant directives at the middle-end level or implementing OpenACC specific code. - device_type clause: Filtering clauses in the GIMPLE IR linked list based on the target device type at compile time using the ACCEL_COMPILER variable. - cache directive: Moving beyond the current no-op behavior by emitting prefetch instructions and/or placing data in low-latency memory using existing GCC prefetch infrastructure and OpenMP memory traits for certain use cases. Additionally, once core tasks are complete, the project will survey newer OpenACC specification features, producing a structured list of unimplemented functionality with prioritized implementations and documentation to ease future contributions. Deliverables include upstream patches for each directive/clause with test suite additions for C, C++, and Fortran, and documentation updates to the relevant documentation files.