Add Stable Diffusion demo

[akhilg-nv]

No description provided.

[akhilg-nv]

Tested for accuracy with pytest tests/test_diffusion.py and visual inspection of outputs compared to references on set seed and prompt. Note that I think the current test setup might need some tweaks to properly enable in CI, and test-specific dependencies could be separated out from the example requirements possibly.

Perf now roughly matches oss/demo/Diffusion on my RTX-4070 Ti. Note that the e2e latency can be marginally improved by removing stream synchronize calls while maintaining correctness, though this would make the per-component timing less accurate.

Tripy:
|-----------------|--------------|
|     Module      |   Latency    |
|-----------------|--------------|
|      CLIP       |      2.82 ms |
|    UNet x 50    |   1169.86 ms |
|     VAE-Dec     |     39.18 ms |
|-----------------|--------------|
|    Pipeline     |   1211.85 ms |
|-----------------|--------------|
Throughput: 0.86 image/s

OSS demoDiffusion:
|-----------------|--------------|                                                                                                                              
|     Module      |   Latency    |                                                                                                                              
|-----------------|--------------|                                                                                                                              
|      CLIP       |      1.80 ms |                                                                                                                              
|    UNet x 50    |   1106.67 ms |                                                                                                                              
|     VAE-Dec     |     29.57 ms |                                                                                                                              
|-----------------|--------------|                                                                                                                              
|    Pipeline     |   1138.24 ms |                                                                                                                              
|-----------------|--------------| 
Throughput: 0.88 image/s