Enhance SOGKSpace and GPU implementation with direct k-space mode and timing metrics

- Updated SOGKSpace class in sog.h to include a new direct k-space mode with associated parameters and methods for enumeration and computation.
- Introduced timing events in SogGpuState for various GPU phases (spread, FFT forward, k-space, FFT inverse, gather) to measure performance.
- Modified sog_gpu.cu to record timing for each phase of the computation and report average times every N steps.
- Clarified B-spline order documentation in sog_spline.h to differentiate between legacy and generalized B-spline orders.

Author: Johnsonjz