Misleading LazyTensor internal dimension management

[bcharlier]

Internal dimension equals to 1 does not appear in the LazyTensor.shape attribute

import numpy as np
from pykeops.numpy import LazyTensor

# work as expected
print(LazyTensor(np.ones((1,10, 2))))
# KeOps LazyTensor
#    formula: Var(0,2,1)
#    shape: (1, 10, 2)

# strange...
print(LazyTensor(np.ones((1,10, 1))))
# KeOps LazyTensor
#   formula: Var(0,1,1)
#   shape: (1, 10)

A better choice should be shape: (1, 10, 1). It yields to strange expression :

print(LazyTensor(np.ones((1,10, 2))).sum(-1).sum(-1))
# KeOps LazyTensor
#    formula: Sum(Sum(Var(0,2,1)))
#    shape: (1, 10)

Not coherent with

print(LazyTensor(np.ones((1,10, 2))).sum(-1).sum(1).shape)
# (1, 1)