Add projection dim to text and vision model configs for CLIPVisionModelWithProjection and CLIPTextModelWithProjection support

The default projection_dim is 512 which will throw an error when loading weights for

```py
from transformers import CLIPVisionModelWithProjection
CLIPVisionModelWithProjection.from_pretrained('laion/CLIP-ViT-H-14-laion2B-s32B-b79K')
```

or

```py
from transformers import CLIPTextModelWithProjection
CLIPTextModelWithProjection.from_pretrained('laion/CLIP-ViT-H-14-laion2B-s32B-b79K')
```

Loading CLIPModel will not throw an error because it uses the `projection_dim` on the top level of the config.

```py
from transformers import CLIPModel
CLIPModel.from_pretrained('laion/CLIP-ViT-H-14-laion2B-s32B-b79K')
```

config.json

@@ -59,6 +59,7 @@
     "pad_token_id": 1,
     "prefix": null,
     "problem_type": null,
+    "projection_dim": 1024,
     "pruned_heads": {},
     "remove_invalid_values": false,
     "repetition_penalty": 1.0,
@@ -85,7 +86,8 @@
     "hidden_size": 1024,
     "intermediate_size": 4096,
     "num_attention_heads": 16,
-    "num_hidden_layers": 24
+    "num_hidden_layers": 24,
+    "projection_dim": 1024
   },
   "torch_dtype": "float32",
   "transformers_version": null,
@@ -144,6 +146,7 @@
     "patch_size": 14,
     "prefix": null,
     "problem_type": null,
+    "projection_dim": 1024,
     "pruned_heads": {},
     "remove_invalid_values": false,
     "repetition_penalty": 1.0,
@@ -170,6 +173,7 @@
     "intermediate_size": 5120,
     "num_attention_heads": 16,
     "num_hidden_layers": 32,
-    "patch_size": 14
+    "patch_size": 14,
+    "projection_dim": 1024
   }
 }