[Dynamo] InternalTorchDynamoError when running dynamo.explain on SparseArch

[bertmaher]

This is a copy of internal task T136878330 for tracking.

N2743445 shows dynamo.explain failing with

Traceback (most recent call last):
  File "/mnt/xarfuse/uid-373971/5bdb8da6-seed-nspid4026531836_cgpid12366328-ns-4026531840/torch/_dynamo/variables/user_defined.py", line 71, in call_method
    return super().call_method(tx, args, kwargs)
TypeError: call_method() missing 1 required positional argument: 'kwargs'

I am pretty sure that's due to a simple missing parameter at a call site (D40938415) but even when that change is applied (see N2748186, which is a clone with a locally-built kernel) we see:

Traceback (most recent call last):
  File "/data/users/bertrand/fbsource/buck-out/v2/gen/fbcode/52f25ab8ad19cfcd/bento/kernels/__bento_kernel_pyper__/bento_kernel_pyper#link-tree/torch/_dynamo/variables/tensor.py", line 344, in create
    raise AssertionError(
AssertionError: torch.* op returned non-Tensor bool call_function <built-in method _profiler_enabled of PyCapsule object at 0x7f8a30b7ffc0>

[voznesenskym]

.explain just runs regular dynamo in "eager" mode, there is nothing special in .explain, anyone is free to take this.

As the kwargs bug is fixed, the real bug is around support for _profiler_enabled