Bugs with shear in realign and spm_imatrix

[dagra]

Hi,
while i was doing some experiments with realign with more parameters than the default and with spm_imatrix i found out the following problems:

1. even though realign.py can compute more parameters than the default 6 rigid body, it returns only the first 6 and i think the problem is in this line where 6 is hardcoded, but it should be dependent on lkp, i.e. sess_rp[..., :len(self.lkp)]

2. spm_imatrix seems incorrect and the following test case identifies the issues

   def test_spm_imatrix_with_shear():
       params = np.asarray([-3.11811347, 3.42120573, -3.67295542,  # translation
                            -0.74295001, 0.20452795, -0.98378122,  # rotation
                            1., 1., 1.,  # zoom
                            1.2, 3., 0.])  # shear
       np.testing.assert_array_almost_equal(spm_imatrix(spm_matrix(params)),
                                            params)
   

   running the above test, IndexError: index 3 is out of bounds for axis 1 with size 3 is raised in this line. Comparing it with spm's spm_imatrix i think it came up from the different indexing in matlab and it should be p[5] = np.arctan2(-rang(R1[1, 0]), rang(-R1[2, 0] / R[0, 2])). After correcting this, the test fails with

   E       AssertionError: 
   E       Arrays are not almost equal to 6 decimals
   E       
   E       (mismatch 41.6666666667%)
   E        x: array([-3.118113,  3.421206, -3.672955,  0.      ,  1.570796, -1.934645,
   E               1.      ,  1.      ,  1.      ,  0.      ,  0.      ,  0.      ])
   E        y: array([-3.118113,  3.421206, -3.672955, -0.74295 ,  0.204528, -0.983781,
   E               1.      ,  1.      ,  1.      ,  1.2     ,  3.      ,  0.      ])
   

   The reason of this error is in this line and the different flattening method where numpy.ravel by default concatenates the rows, while in matlab the columns. So, two possible solutions (i'm not sure which one is preferable) would be

   1. p[9:12] = C.ravel(order='F')[[3, 6, 7]]
   2. p[9:12] = C.T.ravel()[[3, 6, 7]]

If you agree with these issues and corresponding solutions, i can open a PR.