✨ enhance dataset.read_table(...) method with type normalisation

[Marigold]

Motivation

Historically, we've been using function dataset["my_table"] to access table from a dataset. Recently, a new helper method dataset.read_table(reset_index: bool = False) has been added that lets us read the table with reset index which is significantly faster for large dimensional datasets.

Concept

We could add more functionality to read_table and make it de facto standard to read tables. These could be:

• Retype all columns to "standard" types (e.g. uint8 -> int64, Float16 -> float64) and categorical to string type
• underscore column names etc. (see .format method)
  • Should already be in snake case by the time we're talking tables

[larsyencken]

We think type standardising would remove some common footguns for data folk:

• categorical → string
• Float* -> Float64
• Int* -> Int64

Maybe this should live in the repack module.

It would be nice to turn this on by default.