iris.analysis.PERCENTILE fails with TypeError in scipy mquantiles

[munslowa]

Collapsing a 3 dimensional cube (time, grid_lat, grid_lon) over 'time' using the aggregator PERCENTILES results in a TypeError being reported from mquantiles.

raise TypeError("Array should be 2D at most !")

[munslowa]

Example script to reproduce problem:

import iris

dir = '/home/andrew/git/iris-test-data/test_data/NetCDF/global/xyt/'
file = 'SMALL_hires_wind_u_for_ipcc4.nc'

cube = iris.load_cube(dir+file)

print cube

result = cube.collapsed('time', iris.analysis.PERCENTILE, percent=90)

print result

[ajdawson]

Fixed by #625.