[REQ] Support for 128bit integers

[mmacklin]

Description

Add support for 128bit integer types (__int128, __uint128).

Context

Semantic ids in NVIDIA Replicator use 128bit datatypes that would be useful to process in Warp.

[shi-eric]

@Jean-Francois-Lafleche you made this feature request through Miles, right? Can you let us know about how we should prioritize this, e.g. do you have a workaround you're happy with and this is nice to have, or would having this soon really help your use case? Some more information about requirements would also help.

[daedalus5]

For example, do you need full support for 128 bit integer types (meaning these would be as fully featured as any other Warp type), or do you only need to use certain Warp primitives?

[Jean-Francois-Lafleche]

Lack of 128bit support isn't blocking any work that I'm aware of. We have some workarounds using numpy that aren't ideal but that work. The most common workload manipulating these IDs is filtering - so given an array of IDs, we might want to select a subset based on some evaluated criteria and identify the matching array indices to filter a secondary array (eg. bounding box dimensions and transforms)