Pass headers to litapi

Nobody has claimed this yet.

## 🚀 Feature

Being able to access headers in the litapi stages (decode, predict, etc.).

### Motivation

- Tracing - user agents, region and similar in distributed logging across stages (decode, predict, etc.)
- Model tweaks (different model, different thresholds, etc.) based on implicit headers (like user agent).

### Pitch

There is benefit to accessing headers natively in litapi that would provide more flexibility in the model inference and execution since headers are being send implicitly by clients and in some cases they can differ the way predict should be executed.

### Alternatives

Currently, we need to repack the request body if we want to add header value to the request object ending up in the litapi.
Or, caller must pass them in body.

### Additional context

https://github.com/Lightning-AI/LitServe/issues/566

No file or test is named. Start by tracing how requests enter litapi and reach the decode and predict stages, then identify where headers are currently discarded. Done means stage code can access the incoming headers without requiring callers to repack them into the request body, with behavior covered by relevant tests.

Written by the indexing model from the issue text.