Fix distributedType -> distributed_type typo in downcast_bf16 guard (#4149)
The downcast_bf16 validation in Accelerator.__init__ references
`self.state.distributedType`, which does not exist -- the attribute is
`distributed_type` (used on the line above at 520 and throughout state.py).
When mixed_precision != "bf16" and downcast_bfloat is set, the guard is
reached and raises `AttributeError: ... has no attribute 'distributedType'`
instead of the intended, clear ValueError.
Co-authored-by: uttam12331 <uttam12331@users.noreply.github.com>