BUG: Reindex data if plotting time/period index (GH3601)

[dieterv77]

Just wanted to push this fix to see if it looked OK and let the Travis build start.

If it looks ok, i will add something in the release notes and try to write a proper test.

[dieterv77]

Added a test and update release notes. squashed into one commit

[dieterv77]

@jreback, ok to merge this?

[jreback]

@dieterv77 looks good to me

[ghost]

Since the datetimeindex supplies the labels here, why it it mandatory
to plot it in sorted order? at the very least, what about ascending/descending order?

Hardcoding a sort fixes one problem (GH3061), but creates another.
Previously, the user could opt-in (by sorting himself), but now he can't opt-out.
This looks like a regression to me.

Also, @dieterv77 , please mark issues by the miletsone they go into, if you're merging
your own PRs.