Switch storage strategy

[memorycode]

Proposed changes

This PR completely changes the internal storage layout to optimize queries. The goal is to be able to iterate the entities in each archetype in a cache friendly way and with as little indirection as possible

At its core, an archetype stores data as a map from a component -> map from an entity -> component data (fields in the code). However, we assign contiguous indices to each component and to each entity so that the maps are effectively arrays.

All of the supporting structures in an archetype are to make this densely packed layout possible. For instance, idToIndex maps a component ID to its index in fields.

Performance

Iterating ~7,400 entities with 128 archetypes is 7.5x faster. More synthetic benchmarks show 10-15x improvement.

Next steps

There are plenty of optimizations that can still be made. Some will come at the expense of readability and have been left out of this implementation to keep the focus clear, while others were omitted simply due to time constraints and the desire to keep this PR as small as possible.

The most obvious and immediate follow-up is to implement an archetype graph so that insertions and removals can be much faster.

Thanks to @Ukendio for guidance and @SanderMertens for the great articles upon which this implementation is based on.

[Ukendio]

lgtm