[BUG] `Head.build` raises `TypeError: 'NoneType' object is not callable` when body lacks `output_size`

Bug description

PR #802 (commit ab7207cf, "Sec pic fix") changed the default sentinel in transformers4rec/torch/model/base.py from lambda: None to None:

https://github.com/NVIDIA-Merlin/Transformers4Rec/blob/8bf122f5dcb39feecfc6dabde734d79c2d1c4380/transformers4rec/torch/model/base.py#L288

if not getattr(self.body, "output_size", None)():
    raise ValueError(
        "Can't infer output-size of the body, please provide  "
        "a `Block` with a output-size. You can wrap any torch.Module in a Block."
    )

When self.body does not have an output_size attribute, getattr(..., None) returns None, and the trailing () then evaluates None(), which raises TypeError: 'NoneType' object is not callable.

The original lambda: None sentinel was callable and returned a falsy value, so the intended raise ValueError(...) path below was actually reached. The current code path short-circuits with the wrong exception type, making the error message unactionable.

This is not caught by tests because all fixtures provide bodies with output_size.

Steps/Code to reproduce bug

class Body:
    pass

body = Body()
getattr(body, "output_size", None)()
# TypeError: 'NoneType' object is not callable

Versus the pre-#802 behavior:

getattr(body, "output_size", lambda: None)()
# None   (falsy, falls through to `raise ValueError(...)` in Head.build)

Expected behavior

When the body lacks output_size, Head.build should raise the original ValueError("Can't infer output-size of the body ...") so the user gets a clear, actionable message.

Environment details

• Transformers4Rec: main @ 8bf122f5 (regression introduced by PR #802 / commit ab7207cf)
• Python: any

Additional context

Two equivalent minimal fixes:

Option A — restore the callable sentinel:

if not getattr(self.body, "output_size", lambda: None)():

Option B — explicit two-step check (slightly more readable):

output_size_fn = getattr(self.body, "output_size", None)
if output_size_fn is None or not output_size_fn():
    raise ValueError(...)

Option A preserves behavior byte-for-byte. Happy to send a PR.

Start in transformers4rec/torch/model/base.py at Head.build and reproduce the failure with a body that lacks output_size. Verify that the missing attribute reaches the intended ValueError path with its actionable message, and add regression coverage for this case alongside the existing model tests.