Init grid parallelisations

[wilfchun]

This pull request significantly improves the parallel performance and thread safety of the nInitGridAndCalcStillWaterLevel function in src/init_grid.cpp by redesigning how accumulators are handled in OpenMP parallel regions. The main change is the introduction of cache-aligned, thread-local accumulators to eliminate false sharing and reduce synchronization overhead, resulting in more efficient and scalable parallel execution.

Parallelization and performance improvements:

• Introduced a cache-aligned ThreadLocalAccumulators struct and allocated one per thread to prevent false sharing and synchronization bottlenecks during parallel accumulation. This replaces OpenMP reduction clauses with thread-local storage, improving scalability on multi-core systems.
• Updated the parallel loop to use these thread-local accumulators, ensuring each thread writes to its own cache line and removing the need for reduction clauses. [1] [2]
• After the parallel region, combined the results from all thread-local accumulators in a sequential pass, which is efficient compared to the parallel workload.

Correctness and code clarity:

• Changed all per-cell accumulations within the parallel region to use the thread-local accumulators instead of global variables, ensuring thread safety.
• Added detailed comments explaining the motivation for the changes and the reasoning behind cache alignment and accumulator design. [1] [2]

Minor code cleanup:

• Fixed a typo in a commented-out variable name (pdRaster allocation).