Add stable unCLIP variations and VAE slicing

[fboulnois]

Adds stable unCLIP variations which creates different versions of an existing image with a text prompt:

./build.sh run --model 'stabilityai/stable-diffusion-2-1-unclip-small' \
  --image parakeet_eating_spaghetti.png \
  --prompt 'An impressionist painting of a parakeet eating spaghetti in the desert'

Also adds VAE slicing to lower memory use when generating batches of images:

./build.sh run --samples 2 --iters 2 \
  --skip --vae-slicing --xformers-memory-efficient-attention \
  --prompt "abstract art"

Other changes:

• Update torch to 2.0.0, transformers to 4.28.1, diffusers to 0.15.0, and xformers to 0.0.17
• Fix issue with invalid filenames and add test (resolves #77)
• Temporarily suppress UserWarning due to deprecated parameters in diffusers pipelines
• Add tests for VAE slicing and unCLIP variations
• Add VAE slicing option and an example for unCLIP variations to docs