BUG: unwanted conversions of timedelta dtypes when in a mixed datetimelike frame (GH7778)

[jreback]

closes #7778

TST: tests for internals/as_matrix() for all dtypes (including categoricals)

[jreback]

cc @JanSchulz

I fixed up doing something like

df.values when you have a categorical (or 2) as a dtype.

[jreback]

@cpcloud

should

df = DataFrame({'bool' : [True, False], 'int' : [1, 2] })
df.values

be object or int? (I have it as object now).

[cpcloud]

hm int more useful, but object more consistent with the fact that numpy separates bool from int, i think int. is this an api breaking change?

[jreback]

I think it was object before
let me see

[jreback]

@cpcloud only took the usual 5x iterations for make the 1.6.1 stuff work :<

[cpcloud]

😡 on 1.6