computeTemeToPseudoFixedMatrix may have problems near leap seconds.

The three tests below all fail if you change date from `June 29, 2015 12:00:00 UTC` to `June 30, 2015 12:00:00 UTC` (which is near a leap second). The problem is that either these tests simply don't work if crossing a leap second (and hardcoding the dates as done in #2845 is the actual solution), or there is a bug somewhere that needs to be fixed. We should determine the exact cause of the failure and close this if the tests were bad, or fix the bug if it's our computation.

```
computeTemeToPseudoFixedMatrix works before noon
computeTemeToPseudoFixedMatrix works after noon
computeTemeToPseudoFixedMatrix works with a result parameter
```

Start with the three tests named in the issue: “computeTemeToPseudoFixedMatrix works before noon,” “works after noon,” and “works with a result parameter.” Compare their behavior on June 29 and June 30, 2015, then determine whether the failure is caused by invalid test dates near the leap second or by the computation; done means either correcting the tests or fixing the computation and restoring passing tests.

Written by the indexing model from the issue text.