Add `Dataset.get_subhypergraph()` for self-contained sub-hypergraph sampling

[tizianocitro]

Dataset.__getitem__ currently returns an HData with global IDs and no features, designed for efficient batching in DataLoader.collate. This means dataset[0] returns something that isn't directly usable for inspection or standalone computation (e.g., the hyperedge_index has global IDs but x is empty).

I want to add a get_subhypergraph(index: int | List[int]) method that returns a fully self-contained HData:

• hyperedge_index remapped to 0-based local IDs.
• x subset to the nodes in the subgraph.
• y subset to the hyperedges in the subgraph.
• hyperedge_attr subset if present.

This keeps getitem fast and minimal (global IDs only, used by DataLoader), while giving users a clean way to extract and inspect a usable subgraph.

For DataLoader (fast, global IDs, no features):

dataloader = DataLoader(dataset, batch_size=16)

For inspection/standalone use (self-contained, 0-based):

subgraph = dataset.get_subgraph(0)
subgraph.x[subgraph.hyperedge_index[0]]  # works correctly

Notes:

• Reuse the existing __sample_hyperedge_index to filter incidences.
• Remap to 0-based using to_0based_ids (already available in hyperbench/utils/data_utils.py)
• Subsets x, y, hyperedge_attr from self.hdata using the sampled global IDs.

Possible implementation

def get_subhypergraph(self, index: int | List[int]):
        sampled_node_ids_list = self.__get_node_ids_to_sample(index)
        self.__validate_node_ids(sampled_node_ids_list)

        sampled_hyperedge_index, sampled_node_ids, sampled_hyperedge_ids = (
            self.__sample_hyperedge_index(sampled_node_ids_list)
        )

        new_x = self.hdata.x[sampled_node_ids]
        new_y = self.hdata.y[sampled_hyperedge_ids]

        new_edge_attr = None
        if self.hdata.hyperedge_attr is not None and len(sampled_hyperedge_ids) > 0:
            new_edge_attr = self.hdata.hyperedge_attr[sampled_hyperedge_ids]

        return HData(
            x=new_x,
            hyperedge_index=sampled_hyperedge_index,
            hyperedge_attr=new_edge_attr,
            num_nodes=len(sampled_node_ids),
            num_hyperedges=len(sampled_hyperedge_ids),
            y=new_y,
        )