transpose dimensions for consistency

[gbrencher]

Variables in our output xr dataset mostly have dimensions (time, y, x). A few variables (deltaGamma, snow_index, and snow_depth) have dimensions (y, x, time). This isn't ideal because some functions, like ds.rio.reproject, expect (time, y, x). The dims can be easily transposed later on, but that requires loading the whole array into memory, which you might not want to do for a big ds. Would be ideal to transpose during processing.