#3429 demonstrated that roundToNearestMinutes() did not produce expected behaviour. This PR makes a few changes so that roundToNearestMinutes() fixes that misbehaviour...

• Previously, there was an inexplicable addedMinutes variable which added minutes on to the result. There was no obvious reason for this, and it could only ever cause the result to go wrong: indeed, this seemed to be the chief culprit in the case of roundToNearestMinutes producing unexpected results #3429. I have removed this variable
• Fixing this behaviour caused the tests to fail. This was because the tests had been engineered to expect the wrong results. I fixed this, and while I was there, removed duplicate tests and replaced test inputs with boundary cases
• Fixing the tests to expect what they ought to expect and putting in boundary cases caused further test failures. This in turn was partly because there was no account made of the milliseconds field of the input. Now, milliseconds are accounted for, which allowed some more of the tests to pass.
• Finally, the remaining issue was that the rounding method was expected to default to round, but in fact it was defaulting to trunc. I have fixed this as well.

Now, all tests pass, and I have manually checked that this should fix #3429