[Python][ML] Add device param to RDataLoader's .as_torch()

[siliataider]

This Pull request:

Changes or fixes:

as_torch() now supports an optional device argument, allowing batches to be placed directly on the target device without a separate .to(device) call after the iteration:

train = ROOT.Experimental.RDataLoader(df, ...)

for x, y in train.as_torch(device="cuda"):
   ...

Proposed as an alternative to

for x, y in train.as_torch():
    x = x.to("cuda")
    y = y.to("cuda")
    ...

The implementation directly passes device to torch.as_tensor(device) which handles the transfer. Default is None.

[github-actions]

Test Results

    22 files      22 suites   3d 7h 46m 56s ⏱️
 3 845 tests  3 844 ✅  1 💤 0 ❌
76 827 runs  76 809 ✅ 18 💤 0 ❌

Results for commit 1cdfa0f.

♻️ This comment has been updated with latest results.

[siliataider]

Failures unrelated to the changes of this PR, will be fixed by root-project/root-ci-images#112

[vepadulano]

LGTM thanks!

[siliataider]

@dpiparo This PR is ready to merge!