Patch Pandas ExtensionArrayFormatter to handle ndims > 1

[BryanCutler]

Pandas formatting has been failing in TensorArray for datetime64 values and float values with ndims > 1. This patches the ExtensionArrayFormatter to handle higher dimension formatting until a proper fix exists upstream.

Closes #151

[BryanCutler]

@frreiss I'm not crazy about patching like this, but since we can't display TensorArrays with float values of ndim > 1, that is a big limitation. I'm still working on the upstream fix, but waiting on that might prevent us from upgrading Pandas for a while. wdyt?

[BryanCutler]

Still some kinks to work out with text spacing in the tests, I think..

[BryanCutler]

I think I got all the kinks worked out. I had to add some special handling for pandas 1.0.x, but that can all be removed once we bump the minimum version.

[frreiss]

LGTM. You're right that this sort of thing is to be avoided if possible, but I don't see how we an avoid this and still have compatibility with Pandas versions between 1.0 and 1.2.1.

The main risk is that the patched function might break some other code in the user's environment that doesn't use Text Extensions for Pandas at all. That type of crash would be a really nasty one to track down the root cause of.

Have you run the full regression suite of Pandas itself with this patch in place?

[BryanCutler]

@frreiss I added a check to use the original method if values aren't from a TensorArray, added the env var to switch it off completely, and added an upper bound to stop patching at Pandas 1.3,0+. I'm hopeful I'll be able to get a fix in by then :) If not we can just bump that version up. I'm going to go ahead and merge this and we can start using newer pandas versions.

[BryanCutler]

the leading space is because that option defaults to true, we might want to look into changing that for tensors at least