Fix return type annotation of PyRanges.__new__ to enable method chaining in static type checkers

[adamgayoso]

Currently, PyRanges.__new__ is annotated as returning pr.PyRanges | pd.DataFrame. When a user instantiates a PyRanges object:

import pyranges1 as pr
import pandas as pd

df = pd.DataFrame({"Chromosome": ["chr1"], "Start": [0], "End": [100]})
gr = pr.PyRanges(df)

# Type checker error!
# "Item 'DataFrame' of 'PyRanges | DataFrame' has no attribute 'join_overlaps'"
result = gr.join_overlaps(other_gr)

Because PyRanges inherits from DataFrame, any function expecting a DataFrame already accepts a PyRanges instance without needing | pd.DataFrame in the annotation.

[endrebak]

I will have a look

[endrebak]

I do not remember why I wrote it like this. I wonder if this was due to the return type sometimes actually being pd.DataFrame (when Chromosome, Start, or End is missing). This is actually bad design. I should throw an Error instead.

I think I thought that it would break LSP if I did not support this but I was confused. Written before I had LLMs to discuss with. Will check that tests pass and do a minor version bump also.

[endrebak]

Ok, so I am getting back into the groove and it seems like we intentionally want it to be PyRanges | DataFrame so that e.g.

gr.drop("Chromosome")

returns a DataFrame.

[endrebak]

See #171

[adamgayoso]

Thanks, #171 looks good to me.

[endrebak]

Type checks fail. Will need to find the time to work on this. Perhaps Sunday?