Zero-copy support in BufferedTrajectory (#96)

- Use contiguous arrays to hold various Timestep arrays for full buffer size
- Use a simple ring buffer of above arrays to hold full Timesteps
- Load and return Timesteps directly into / from the buffer without copy
- A new MultiArrayBuffer class to hold multiple np arrays in one mem block
  - individual np arrays 64-byte aligned
  - support for using shared memory (across parallel workers)
  - caching support on parallel workers to avoid re-attach every time
- Cleanup handling of the buffer across disconnect / connect, shutdown
- Universe config option in UI to enable use of shared memory (default True)

Author: PardhavMaradani

CHANGELOG.md

@@ -30,6 +30,7 @@ The rules for this file:
 <!-- New added features -->
 
 - Parallelization improvements (PR #93)
+- Zero-copy support in BufferedTrajectory (PR #96)
 
 ### Fixed