Refactor: do not route vals_ methods through GT object

[machow]

Currently, functions like vals.fmt_number() work as follows:

• The input list or value is converted to a pandas DataFrame
• The DataFrame is wrapped in GT
• The corresponding fmt_ method is run
• The formatted values are extracted as a list

This is problematic for two reasons:

• It creates a hard dependency on pandas
• It feels a bit circuitous to wrap in GT, run, then take out

Removing use of GT

Fortunately, it seems that GT.fmt_* methods don't make much use of the GT object. Uses seem to be largely:

• Getting a default locale setting
• Passing in the DataFrame to dispatch on for is_na() checks
• At the very end, attaching to GT._formatters, and returning a new GT object

It seems like we could pretty easily run vals.fmt_* funcs straight over formatter logic. This would let us remove the pandas dependency.

Alternatively..

An alternative would be for us to support GT(SomeInBuiltFrame). GT's core logic does not require a very beefed up DataFrame implementation. We could get away with a class that's essentially a dictionary of same-length lists. This might actually be easier, since we wouldn't need to refactor the formatter code.

[machow]

Alternatively alternatively...

We could support inputting a pandas series or polars series as the input values. We have the machinery to detect these, and at that point do pandas/polars specific things w/o creating any dependencies.