Add xfail test for df.astype(pa.DictionaryType)

[NickCrews]

Demonstration of
#2187

[maartenbreddels]

So, it works for some cases, because our dtype wrapper compared true to str when it's an encoded type that maps to str.
Next problem is, that because of the expression system, we would need to add the types passed in to the variables, because a repr(arrow_dtype) cannot be evaluated again.
This would need to use a similar ideas as #2089 where we don't add types multiple times.

For the rest, I'll first wait for your response to #2187 (comment)