Refactoring ideas

[regDaniel]

Just a few ideas that came up during #114 :

• Split Ceilochunk into dedicated classes for slices, groups and layers, possibly all inheriting from the same base class.
• Allow for metadata assignment by creating a dedicated DataFrame class which inherits from pandas.DataFrame or holds a DataFrame and additional attributes that can be validated (a pydantic data class)
• This could go hand in hand with using polars to speed up operations on dataframes (Could polars help speed up ampycloud even more ? #96)