BUG: array raveled but not reshaped in all cases

[RobinFiveWords]

• closes BUG: Transpose construction/block error with certain dtypes #16362
• tests added / passed
• passes git diff upstream/master --name-only -- '*.py' | flake8 --diff
• whatsnew entry

[RobinFiveWords]

Sorry about that, not appropriately tested.

[jreback]

@RobinFiveWords you can reopen and push tests when you are ready

[RobinFiveWords]

@jreback Thanks, more that my code was bad the first time, didn't understand the full scope of how the function was used. When I update my branch, can I reopen this PR with that new code?

I haven't figured out an elegant way to have maybe_infer_to_datetimelike do exactly what its docstring says, specifically returning the input unchanged. Not sure how important that is, though. I imagine the lib.infer* functions (not sure where in _libs/src to look for them) require a 1-dimensional np.array, and if our input was of higher dimension, we need to reshape exactly once and only after we're done trying to convert. The simplest change, as I see it, is to allow a just-reshaped np.array to be returned in all cases. core/frame.py and core/categorical.py are each going to coerce to np.array in the very next step anyway.

If something readily comes to mind that I'm missing, please let me know!