Add while loop batching

[tchaton]

This PR adds an example of inference as a loop for stable diffusion.

Computed on RTX 3090:

Type	Max Batch Size	Number of users (locust)	Average (ms)
Grouped Batch	6	6	8868
Streamed Batch (naive)	6	6	9442
Streamed Batch (global state)	6	6	8584

And the following benchmarks on T4.

Type	Max Batch Size	Number of users (locust)	Average (ms)	Number of requests
Grouped Batch	1	1	5061	1353
Grouped Batch	2	2	11393	1599