moon location shouldn't be static for split exposures

[sbailey]

When simulating split exposures (i.e. multiple exposures of the same tile in a row), all exposures receive the same MOONALT and MOONSEP even though the moon has moved from one exposure to another. Additionally, they all have the same MOONFRAC for the full night.

The underlying cached ephemeris is on a 1 hour grid with cubic interpolation in (ra, cos(dec), sin(dec)), but it appears that we are sampling that only once for the beginning of the exposure sequence rather than updating it for each exposure.

From David Schlegel:

I think this is a desisurvey issue rather than a surveysim issue, but feel free to move it if needed.