speedup collection struct

[vagabondkjo]

This PR speeds up the collection struct.

1. Some implemented ideas

a. Add new element

• improve finding location to add an element into the element list of a POV entry using binary search tree instead of the old linked list approach.
• try to make balanced search tree if we got number of interations exceed a dynamic number (threshold). Ideally, build this once, and it should help to speedup finding procedure later.

b. Optimize accessing to memory

• fit encoded 32 occupation flags per read for 32 POV entries instead of currently doing per check per read the flags buffer.

c. Optimize checking occupation flag

• use trailing-zeros/leading-zeros intrinsics to jump-out/ignore some bits quickly

d. Optimize cleanup function

• use smaller buffer and do resetting/setting memory at least as possible.
• don't need to rebuild/copy elements buffer, because we only need to modify element's povIndex property
• quick stop by checking population info

2. Profile

• the performance has been improved around 20-30%.

[philippwerner]

Thanks, this is awesome work. Please revise according to my comments.

Please also consider to add some of the "Some implemented ideas" mentioned above in comments in the source code.

[vagabondkjo]

@philippwerner thanks and addressed your comments.

[philippwerner]

Great, thanks. Time for merging.