Make a unit test for the bug with TT2000 conversions.

[bryan-harter]

This is in reference to PR #292 and issue #233

Let me recap this just for my later knowledge:

1. The chunk of code I modified only does anything if there is data before 1972 in the array (and a "NaT" counts as before 1972)
2. For some reason (that I'm still unclear about), we subtract 1 second from the array, and then add 1 second back in later
3. The code that checks if we go over 60 seconds when we add 1 second back is supposed to +1 to the minutes, and subtract 60 seconds from the seconds. Otherwise, it adds +0 to the minutes (i.e. keeps it the same). Normal time handling stuff.
4. INSTEAD, that code sets the minutes EQUAL to 0 or 1, NOT +0 or 1.
5. It also rounded when I believe we want the np.floor() function, but that is relatively minor in comparison.
6. As a result, the minutes keep getting reset down between 0 and 1. And then the only time there is a significant change is when the hour changes.

The functions which is affects are also INCREDIBLY dense, with variables names like "l" and "i", with basically no comments. It took me about 6 hours to figure out what was going on, just to change that one line. So ideally, better variable names and documentation will also be address by this ticket.