Configure GPU execution from simple.in

[krystophny]

Summary

• Configure GPU mode, backend, integrator, Landreman controls, scheduling, and profiling through the config namelist in simple.in.
• Keep gpu_method = 'auto': native CUDA selects tuned DOPRI; OpenACC selects the supported fast Euler path. Explicit methods remain available where supported.\n- Reduce the GPU build interface to -DSIMPLE_ENABLE_OPENACC=ON or -DSIMPLE_ENABLE_CUDA=ON; defaults provide the remaining settings. OpenACC uses managed memory by default and keeps compiler flags target-scoped so libneo flags do not conflict.\n- Shorten the README and GPU documentation to match the configuration.\n\n## Validation\n\n- Clean GNU build of simple.x.\n- CPU test_gpu_orbit_bench, test_gpu_midpoint_bench, and test_gpu_landreman_segments pass.\n- Legacy SIMPLE_GPU_* variables were set during the segmented test and were ignored; the namelist controlled the run.\n- Final acluster Tesla T4 smoke test: job 21383373, commit 289279b, minimal OpenACC configure, and the three GPU tests using the established libneo/Fortnum checkouts.