DateMath: Use time zone when rounding.

[cbuescher]

Currently rounding in DateMathParser is always done in UTC, even
when another time zone is specified. This is fixed by passing the time zone
down to the rounding logic when it is specified.

Closes #9814

[rjernst]

LGTM

[cbuescher]

On 1.x with 8391b51.

@rjernst I ran into a bit of a mess trying to backport the tests in DateMathParserTests for this PR to the 1.4 branch. The diff between the two branches is quiet big for that test class, so adding just the new lines for this fix is difficult. I tried backporting the the whole test class form 1.x to 1.4, but in that case I had to change a lot of the actual tests to the current behaviour of DateMath on 1.4.
Any suggestions how to procede here? Keep the orginal DateMathParserTests from 1.4 and only try to add the new tests there or try to update the whole thing according to the current state of tests in 1.x?

[rjernst]

If the backport is too complicated, we should not do it. I don't think this is critical enough that it must be in a bugfix release.

[cbuescher]

@rjernst The backport is not difficult, would be nice to have on 1.4. Just the test class changed a lot from 1.4 to 1.x, so the tests will have to be adapted and be slightly different from this PR. Would you mind having another look if I push to my repo before merging?

[rjernst]

@cbuescher Sure, let me know where to look.

[cbuescher]

I pushed the changes agains 1.4 branch to cbuescher@997b8ce. Basically it's the same fix as in 1.x, I only backportet the rounding tests and had to adapt them to the "old" rounding logic on 1.4. (I think you changed the edge inclusions on the 1.x branch)

[rjernst]

@cbuescher +1 to the backport

[cbuescher]

Great, pushed to 1.4 branch with dff19cb