Enable interactive debugging in llm-compressor by skipping self-references in locals (#1617)

If one tries to run interactive debugging session in e.g. VSCode,
LLM-Compressor would crash with the following error:

```
ValueError: Some keys are not used by the HfArgumentParser: ['kwargs', 'local_args']
```

This happens because dictionary `local_args = locals()` contains
`local_args` and `kwargs` as keys (due to different execution context in
debugpy compared to the standard "script mode" with `python
quantize.py`), which HF argument parser in OneShot can't handle. This
doesn't happens when LLM-Compressor is used in a "script mode" with
`python quantize_my_model.py`.

This patch makes it explicit that `local_args` and `kwargs` shouldn't be
in local_args and makes interactive session with LLM-Compressor
possible.

---------

Co-authored-by: Brian Dellabetta <brian-dellabetta@users.noreply.github.com>

Author: eldarkurtic

src/llmcompressor/entrypoints/oneshot.py

@@ -302,8 +302,9 @@ def oneshot(
     """
 
     # pass all args directly into Oneshot
-    local_args = locals()
-    local_args.pop("kwargs")
+    local_args = {
+        k: v for k, v in locals().items() if k not in ("local_args", "kwargs")
+    }
     one_shot = Oneshot(**local_args, **kwargs)
     one_shot()