implement stacey boundary condition for couple_with_injection on GPU, add

[nqdu]

1. Enhanced Fluid Layer Support:

• The FK package now supports fluid layers on top of a stack of elastic layers.

2. Optimized GPU Implementation:

• Direct GPU implementation of Stacey's boundary condition for injections, without copying data back to the CPU.

3. Unified Injection API:

• Introduced a unified API for the injection problem: The main program will call fetch_injection_wavefield before compute_forces_acoustic_forward_calling. In this step the wavefield is correctly read from disk or memory and copied to the GPU if needed.
• For acoustic elements, Tract_axisem/dsm/specfem/fk now save $\dot{\chi}$ and veloc_axisem/dsm/specfem/fk save displ.

[danielpeter]

👍 looks great