* Add separate PArrayTracker for each phase
* Redefine the ResourcePool type to be more flexible & perform compile time checks
* Fix unsigned integer conversion in parray index 
* Count reserved & mapped memory for task in the counter.
* Remove Device::parrayid_toglobalid. 
* Add data task counter.

…switches

…ation-to-c++

…worker-finalization-to-c++

117 optimize task cleanup move worker finalization to c++

* fix: Make crosspy optional
* chore: Build dependency changes
* feature: Port compilation to use newer scikit-build
* feature: Dynamic recompilation of file changes
* fix: MacOS compilation
* fix: CUDA compilation
* feature: Add HIP backend to runahead scheduling (#143)
* fix: Filter out initialization of unsupported CuPy targets in ROCm
* fix: update makefile to new build system
* initial: build C++ stream pool
* style: lint & reformat cython (#146)
* chore: resolve warnings and errors in cython-lint
* chore: update dependencies in pyproject.toml add "dev" extras tag
* fix: restore relative paths (scikit-build-core resolved the error on their end)