Update cord segmentation based on contrast-agnostic model

[jcohenadad]

The current cord segmentation is too small and not representative of the spinal cord CSA (see notably discussions with @vcallot).

The idea would be to:

• re-create the PAM50_cord mask using the contrast-agnostic model
• symmetrize the output
• use that output to update AMU_to_PAM50 registration (Improve registration of AMU WM/GM mask and PAM50 template #41)

This issue should be fixed and merged before working on #41