Optimizations for grid-based dense subdomains

[w1th0utnam3]

• Special handling for "very sparse domains" (domains with less than 5% particles of largest domain) by keeping track of all cubes with at least one edge above the surface threshold
• Compute neighbor lists and densities only for particles inside of the subdomain (ghost particle values will be computed by other subdomains)
• Use flat vectors for serial neighbor list computation instead of nested Vecs