Add presets for CLIP and fix some minor bugs

[james77777778]

This PR:

• Add tests for CLIPBackbone
• Add conversion script for CLIP
• Introduces 8 presets for CLIP from OpenAI and OpenCLIP repo
• Fixes several dtype bugs in CLIP
• Adds support for sublayer and submodel in init_kwargs in run_backbone_test

@divyashreepathihalli

Some results (from conversion script):

Preset	Modeling difference
clip_vit_base_patch16	0.009210587
clip_vit_base_patch32	0.0061130524
clip_vit_large_patch14	0.008178234
clip_vit_large_patch14_336	0.004854679
clip_vit_b_32_laion2b_s34b_b79k	0.02490139
clip_vit_h_14_laion2b_s32b_b79k	0.0014100075
clip_vit_g_14_laion2b_s12b_b42k	1.2397766e-05
clip_vit_bigg_14_laion2b_39b_b160k	1.04904175e-05

Preprocessing difference: 0.095298484

[divyashreepathihalli]

LGTM

[james77777778]

I forgot to register CLIP presets which caused the CI to fail. It should be fixed now.

EDITED:
The GPU CI failure should be irrelevant