streamline vector calc

[tkphd]

Finite differencing should have better performance

• Bug appears in MMSP.grid.cpp in functions laplacian(GRID,...), etc.
• The expected behavior is fast computation of gradients, Laplacians, and divergence using built-in functions.
• The observed behavior in profiling is unusually high residence time in laplacian() and operator()(GRID, vector<int>).

Proposed fix

• Our stock implementation of discrete Laplacian and gradient operators computes the addresses of neighboring cells every time the function is called, despite neighboring coords being static entities.
• Operators then call operator()(GRID, vector<int>), which passes coords through check_boundary before computing the index location.
• Entire process would be much faster by creating a function neighbor_coords which would store a grid of n-dimensional coordinates for high- and low-side neighboring cells, and applying check_boundaries to each stored position vector.
• Discrete vector operators could then call GRID[x][y] instead of GRID(x) to efficiently access stored values, reducing overhead and improving performance.