FluidX3D v3.8 (faster rendering & fixes)

This update continues where the last left off, adding local memory optimization also to graphics_flags_mc() and graphics_rasterize_phi() kernels.

And an interesting bug fix: It's possible that large GPU kernels use too many registers - more variables than can fit into the GPU's SRAM. Kernel execution then fails. This was the case with my graphics_q() Q-criterion isosurface rendering kernel - with local memoy optimization enabled - on old Nvidia Maxwell/Pascal GPUs.
The easy workaround would be to just disable the local memory optimization on these Maxwell/Pascal GPUs. But that would drop rendering framerate by 50%.
Instead I did register occupancy optimization: At any point in the execution timeline, a kernel occupies X registers - this can be traced and counted by hand. In one or more places, X exceeds the SRAM size. Shuffle operations around, re-use previous variables instead of declaring new ones, reduce how many variables reside in registers at once. Voilà, from 152 registers down to 122 registers, just below Maxwell/Pascal's limit of 128 registers. Full rendering performance even on the old hardware, no excuses.

Also some other rendering optimizations, and fixes for Intel Panther Lake and Nvidia GB10.

Improvements

• made graphics_flags_mc() kernel up to 40% faster with local memory optimization
• added local memory optimization also for graphics_rasterize_phi() kernel
• small optimization in marching_cubes()
• dispatch marching-cubes kernels only for (Nx-1)*(Ny-1)*(Nz-1) instead of N
• updated driver install instructions

Bug fixes

• fixed graphics_q() kernel with GRAPHICS_LSQ=8 on older Nvidia GPUs through register optimization
• fixed z-buffer flickering on Intel Panther Lake iGPUs
• disabled zero-copy on Nvidia iGPUs such as GB10 because CL_MEM_USE_HOST_PTR is broken there
• fixed missing allocated_bitmap/zbuffer in Camera move assignment
• fixed missing type cast in OpenCL-Wrapper
• fixed crash/segfault on exit
• removed opencl_unroll_hint again (unsupported on some older GPUs)

Have fun with the software!
-- Moritz