Bug in a network with a generic input connection to a projection weight update

[sebjameswml]

Here, the generic input is plumbed into the weight update component of a projection which has an explicit list of connections. Brahms produces an error like

E_USER: expected data to have 1st dimension of 2
    whilst validating input "out"
    whilst firing EVENT_INIT_CONNECT on GPeTI_to_GPeTI_Synapse_0_weight_update

See attached SpineML project to reproduce the problem more or less as simply as possible.
GI_to_WU_Issue.zip

[sebjameswml]

In the demo, change the connectivity of the recurrent GPeTI to GPeTI connection to "All to All" to make the simulation run.