Make x, y, and z trainable

[michaeldeistler]

Currently, x, y, and z are listed in nodes but are not trainable. After discussing with @jnsbck, we found a few a few initial issues with making those trainable:

• How to compute the length from just compartment centers?
• How to deal with SWC morphologies?
• How to warn users about compartment ends no longer being connected?

Our proposed solution is to make an offset trainable and rely on the user to define the initial x, y, z:

branch.make_trainable("x_offset")
branch.comp(0).make_trainable("y_offset")