Bug in Pandas 1.3.0 block manager breaking TensorArray slicing operations

[frreiss]

Due to some performance optimizations introduced in pandas-dev/pandas#40353, Pandas turns DataFrame.iloc[slice(x, y, z)] into __getitem__((..., slice(x, y, z)) on the ExtensionArray that backs any column defined with an extension type. This bug is breaking indexing on TensorArrays. TensorArray.__getitem__() receives the multidimensional slice spec and returns a 2D slice of the underlying NumPy array instead of the 1D slice that the caller expects to get.

[frreiss]

Opened pandas-dev/pandas#42430 to cover the root cause issue.