.connect() and fully_connect() methods

[michaeldeistler]

Closes #180. FYI @jnsbck @kyralianaka @lappalainenj

Single connections

Following the API of NEURON, this PR implements the following:

net = Network(cells)

pre = net.cell(0).branch(0).comp(0.0)
post = net.cell(1).branch(0).comp(0.0)
pre.connect(post, GlutamateSynapse())

which is equivalent to:

connectivities = [
    [GlutamateSynapse(), [jx.Connection(0, 0, 0.0, 1, 0, 0.0)]]
]

net = Network(cells, connectivities)

Fully (or sparsely) connected layers

net = Network(cells)

pre = net.cell([0, 1, 2])
post = net.cell([3, 4])
pre.fully_connect(post, GlutamateSynapse())

p = 0.2  # Connection probability
pre.sparse_connect(post, p, GlutamateSynapse())

[lappalainenj]

Awesome! Minor suggestions for improvement in inline comments.

[lappalainenj]

I would start with a short paragraph as a comment describing the implementation strategy to scaffold the information for any dev, reviewer, user.

# High-level strategy
# We need to first check if the compartment [branch, cell, network ?] already has a type of this synapse, else we need to register it as a new synapse in a bunch of dictionaries which track synapse parameters, state and meta information [Is this correctly inferred from the code?]. 
# Next, we register the new connection in the synapse dataframe.
# Then, we update synapse parameter and state arrays with the new connection.
# Finally, we update synapse meta information.

[michaeldeistler]

Awesome, copied it almost verbatim into the docstring. And yes, your description is correctly inferred from the code.

[lappalainenj]

Why reversed from 1 to 0 instead of 0 to 1? Cause I would expect that index 0 corresponds to location 0.0

[michaeldeistler]

Yes, indeed, this is unintitive. See #30