HF LLaVa support

@riccardofelluga is already working on this.

Since Sep 19, 2024.

## 🚀 Model / language coverage

The idea is to support LLaVa model from HF. This issue is mainly for tracking the status.

Blocking issues:
- [ ] #735
- [ ] #124

### Minimal Repro

First of all get the `transformers` library with `pip install transformers` then run this script:

```python
import torch
import thunder
from transformers import LlavaForConditionalGeneration

model = LlavaForConditionalGeneration.from_pretrained(
"llava-hf/llava-1.5-7b-hf",
torch_dtype=torch.bfloat16
)
model.to("cuda")

input_ids = torch.randint(1, 32000, (1, 22), device="cuda")
attention_mask = torch.ones((1, 22), dtype=torch.int64, device="cuda")
pixel_values = torch.randn((1, 3, 336, 336), device="cuda")
labels = torch.randint(-100, 32000, (1, 22), device="cuda")

# Setup fake image id
input_ids[0, 0] = 1
input_ids[0, 5] = 32000

model = thunder.jit(model, executors=thunder.get_default_executors())

out = model(input_ids=input_ids, attention_mask=attention_mask, pixel_values=pixel_values, labels=labels)
```