Refactor to make it backend-agnostic, improved GPU execution

[urlicht]

Version up to v1.0.0. Make it backend-agnostic core (supporting both CPU and CUDA), improved GPU execution, added benchmarking tools, added CI.

• 11d205c, e451d21: Refactored registration logic to be backend-agnostic and introduced CPU/CUDA extensions with optional CUDA dependency.
• 29102e6: Removed unnecessary GPU host copies and added true in-place resampling for better performance.
• 34af435, 3726e9f: Fixed CuArray peak-search behavior and reduced allocations via scratch-buffer optimization.
• 095101f, 86b0cdd: Added core unit tests plus synthetic noisy-stack registration tests.
• a4e94dc, c68b125, 0c2162f, 9445820, 6252095, f0c9fa8, e0e794e: Added benchmark suite and commit-compare tooling; improved CUDA benchmark reliability and reporting.
• a5ac633, 7f74312, 816fbc5: Added CI workflow, adjusted CI matrix, and surfaced CI badge in README.
• dc0c075 (plus 399f945): Bumped version to 1.0.0 and updated package author metadata.
• performance improvements including using plan/in-place fft. testing with compare_reg_stack_translate_commits.jl(256x256x64) improvement 13.9% reduction in time from 6252095 to 497f512