Allow Shortcutting Min-max Observer

[kylesayrs]

Purpose

• Speedup runtime of algorithms which require many observations with no averaging such as [GPTQ] Iterative Parameter Updating #863

Changes

• Renamed MovingAverageMinMaxObserver -> MinMaxObserver since moving average is not required to use it
  • Left a deprecated MovingAverageMinMaxObserver in its place
• Shortcut averaging logic by checking self.averaging_constant == 1.0
• Update docstrings, ect.

Testing

• Ran examples/quantization_w4a16/llama3_example.py to completion

from llmcompressor.observers.min_max import MovingAverageMinMaxObserver, MinMaxObserver
observer = MovingAverageMinMaxObserver()
# <stdin>:1: DeprecationWarning: The class name `MovingAverageMinMaxObserver` has been deprecated, please initialize with `MinMaxObserver` in the future
assert isinstance(observer, MinMaxObserver)

[github-actions]

👋 Hi! Thank you for contributing to llm-compressor. Please add the ready label when the PR is ready for review.

[brian-dellabetta]

🚀