Returned "path" of `HTTPReader` and `GDriveReader` diverges

HTTPReader returns the URL for the "path"

https://github.com/pytorch/data/blob/13b574c80e8732744fee6ab9cb7e35b5afc34a3c/torchdata/datapipes/iter/load/online.py#L46

while GDriveReader returns the file name

https://github.com/pytorch/data/blob/13b574c80e8732744fee6ab9cb7e35b5afc34a3c/torchdata/datapipes/iter/load/online.py#L129

Since OnlineReader determines at runtime whether to call the HTTP or GDrive download

https://github.com/pytorch/data/blob/13b574c80e8732744fee6ab9cb7e35b5afc34a3c/torchdata/datapipes/iter/load/online.py#L198-L203

the "path" of the yielded tuples is impossible to predict:

from torchdata.datapipes.iter import IterableWrapper, OnlineReader

dp = IterableWrapper(
    [
        "https://raw.githubusercontent.com/pytorch/data/main/LICENSE",
        "https://drive.google.com/uc?export=download&id=1GO-BHUYRuvzr1Gtp2_fqXRsr9TIeYbhV",
    ]
)
dp = OnlineReader(dp)

for path, _ in dp:
    print(path)

https://raw.githubusercontent.com/pytorch/data/main/LICENSE
torchvision.txt

We should align the two. My vote is out to align based on the file name. Still, returning the URL could also be useful if redirect logic as discussed in https://github.com/pytorch/vision/pull/6060#pullrequestreview-980646299 is added to the HTTPReader.

Start in torchdata/datapipes/iter/load/online.py at HTTPReader, GDriveReader, and OnlineReader. Compare the path values yielded by the two readers and how OnlineReader selects between them, then resolve which path semantics should be consistent. Reproduce the example in the issue and verify that both reader types yield the agreed form.