ModelnnUNetWrapper fails with nnunetv2 >= 2.7: build_network_architecture signature changed #9239

Describe the bug
ModelnnUNetWrapper.__init__ (and therefore get_nnunet_monai_predictor) calls

trainer_class.build_network_architecture(
    configuration_manager.network_arch_class_name,
    configuration_manager.network_arch_init_kwargs,
    configuration_manager.network_arch_init_kwargs_req_import,
    num_input_channels,
    plans_manager.get_label_manager(dataset_json).num_segmentation_heads,
    enable_deep_supervision=False,
)

(monai/apps/nnunet/nnunet_bundle.py, around line 219 on dev). That matches the signature in nnunetv2 2.4–2.6. In nnunetv2 2.7.0 the static method became

build_network_architecture(plans_manager, configuration_manager, num_input_channels, num_output_channels, enable_deep_supervision=True)

so with nnunetv2 2.7.0 or later the positional arguments are shifted and the call fails:

TypeError: nnUNetTrainer.build_network_architecture() got multiple values for argument 'enable_deep_supervision'

get_network_from_nnunet_plans is not affected, because it calls nnunetv2.utilities.get_network_from_plans.get_network_from_plans, whose signature didn't change.

To Reproduce
With nnunetv2==2.8.1 and torch (CPU is enough) installed:

1. Create a model folder <root>/fold_0 with <root>/plans.json (any standard 2d PlainConvUNet plans), <root>/dataset.json, and <root>/nnunet_checkpoint.pth containing {"trainer_name": "nnUNetTrainer", "init_args": {"configuration": "2d"}}.
2. Run ModelnnUNetWrapper(predictor, "<root>/fold_0") (or get_nnunet_monai_predictor("<root>/fold_0")).
3. The TypeError above is raised.

Expected behavior
The wrapper builds the network with any nnunetv2 release that MONAI supports.

Environment
MONAI dev @ 2c54111, nnunetv2 2.8.1 (also checked: the new signature is in 2.7.0 and 2.8.0; 2.4.2–2.6.2 use the old one), torch 2.15.0+cpu, Python 3.12.

Additional context

• No CI job installs nnunetv2, so tests/integration/test_integration_nnunet_bundle.py doesn't catch this.
• A possible fix is to choose the call by signature, e.g. pass plans_manager, configuration_manager, num_input_channels, num_output_channels when the first parameter of build_network_architecture is plans_manager, otherwise use the current arguments. Alternatively, pin nnunetv2<2.7 for the bundle helpers.
• Check nnUNet plans network names against allowed modules #9237 edits the lines just before this call, so whichever merges second may need a small rebase.