Extend summary2.summary_col features

[EBoisseauSierra]

Is your feature request related to a problem? Please describe

I report results of my models via statsmodels.iolib.summary2.summary_col.
However, I have to work around the default implementation to fit my own requirements. These include:

1. Report t-stat instead of standard error under the coefficient,
2. Add the adjusted R² and number of observations,
3. Have a different float level for coefficient and t-stats ('%.4f' vs. `'%.2f').

Describe the solution you'd like

I would like to modify summary_col along the lines of something like this:

  def summary_col(
      results,
+     parentheses: str | None = "std_err",
-     float_format: str ='%.4f',
+     float_format: str | dict[str, str] ='%.4f',
      model_names=(),
      stars=False,
      info_dict=None,
      regressor_order=(),
      drop_omitted=False,
-     include_r2=True,
+     include_r2: bool =True, # adds R^2 only
+     include_adj_r2: bool =True,
+     include_n: bool =False,
  ):

Concretely, this means:

1. parentheses: str | None = "std_err": Choose whether to add stats below the coefficient (cf. None). If so, which one (something along the line of assert parentheses in ["std_err", "t_stat", "z_stat", "proba"]).
2. float_format: str | dict[str, str] ='%.4f',: Pass either one format (current behaviour) that apply to all fields, or something like {"coefficient": "%.4f", "parentheses": "%.5f", "N": "%d"}
3. include_*: Split include_r2 into two flags (one for R², and another for adjusted R²), and add new ones for N (and other fields?)

The features listed above a independent of each other. I'd be happy to raise a PR for each, but I'd prefer to check beforehand whether these would fit your roadmap — and if you have a better way to tackle my issues.

Describe alternatives you have considered

Regarding including adj-R² only and N, I am aware I can use info_dict (what I'm currently doing).
I know as well that I can write my own aggregator method, based on statsmodels.iolib.summary.Summary attributes.

However, I believe the features outlined above could be beneficial to others than me.

Additional context

Currently using statsmodels==0.14.0, and using OLS only.