[data][bug] Dataset.context not being sealed after creation

[raulchen]

Ideally, datasets should capture global DataContext when the dataset is created for the first time.
However, a lot of data code is using DataContext.get_current(), instead of the captured DataContext.
This prevents using multiple datasets with different DataContexts.

#41569 is the first attempt to mitigate this issue. But it only fixes the issue for training jobs, where multiple datasets will be propagated to different SplitCoordinator actors for execution.

If multiple Datasets are running in the same process. This bug still exists.

[raulchen]

After we fix this issue, #40116 should be reverted.

When iterating over a dataset with iter_batches, the execution happens on streaming split coordinator actor. The issue should be fixed by #41569.
However, some API may trigger execution on the local training worker process. For example, to_tf will call schema and trigger local execution. In this case, the DataContext being used is still incorrect.