[ENH] Add Polars Support

[imann128]

Request

Add support for Polars DataFrames alongside the existing pandas implementation,
So users working in Polars-native pipelines can use tsauditor without converting
to pandas first.

Current state

tsauditor is currently pandas-native throughout. Every module assumes:

• a pd.DataFrame input
• a pd.DatetimeIndex as the time index
• pandas-specific operations (.rolling(), .diff(), .groupby(), .iloc[])

What this would involve

At minimum, one of:

• A conversion layer at the scan() entry point that accepts a Polars DataFrame
  and converts to pandas internally before passing to modules
• A parallel Polars-native implementation of each audit module

The conversion-layer approach is simpler and probably the right starting point.
It keeps the core detection logic unchanged while broadening input compatibility.

Notes for contributors

This is a non-trivial addition. A good starting point would be:

1. Opening a discussion comment here on the preferred approach (conversion vs native)
2. Adding polars as an optional dependency in pyproject.toml
3. Implementing and testing the conversion layer in tsauditor/utils/validation.py
   where input validation already lives

If you're interested in taking this on, comment here so we can avoid duplicate work.