Handle Datasource coordinates boundaries (in interpolation)

[jmilloy]

Description
The interpolation may need to behave differently depending on the datasource coordinates boundaries, e.g. for off-center points.

For example, nearest interpolation for "left-aligned" time coordinates should potentially behave differently than nearest interpolation for "centered" time coordinates. should behave differently for time coordinates.

Describe the solution you'd like

The boundaries spec and attribute is already implemented, but using any non-standard boundary raise a NotImplementedError. The interpolators should inspect the boundaries of the source coordinates and adjust their calculation as needed. We still need to develop the spec for some specific needed source coordinate boundaries cases and how the interpolation is affected.

Sample Code

data = np.random.random((100, 100))
coords = Coordinates([np.linspace(0, 10, 100), np.linspace(0, 10, 100)], dims=['lat', 'lon'])
node1 = Array(source=data, coordinates=coords)
node2 = Array(source=data, coordinates=coords, boundary={'lat': [0, 1], 'lon': [0, 1]})
node3 = Array(source=data, coordinates=coords, boundary={'lat': [-1, 0], 'lon': [-1, 0]})

c = Coordinates([np.linspace(1, 2, 20), np.linspace(1, 2, 20)], dims=['lat', 'lon'])

o1 = node1.eval(c)
o2 = node2.eval(c)
o3 = node3.eval(c)