NetX converter is setting num_message_bits incorrectly for recurrent connections.

[tnowotny]

Describe the bug
The NetX converter does not produce a working network if using a cyclic buffer for input and has recurrent connections in the first hidden layer because the num_message_bits are set incorrectly for the recurrent connections (should be set to 0).

To reproduce current behavior
Steps to reproduce the behavior:
This is in the context of work we have done with Jonathan Timcheck at Intel. If you clone
https://github.com/genn-team/ml_genn_netx
and check out the "device" branch.

Also install
https://github.com/genn-team/genn
and
https://github.com/genn-team/ml_genn

Then run
examples/shd.py in ml_genn_netx

you eventually get

pred: [13], labels: [10]
Lava test accuracy: 0.0%

and the screenshot 1 attached below.

Expected behavior
A clear and concise description of what you expected to happen.

Instead one should get:

pred: [10], labels: [10]
Lava test accuracy: 100.0%

and the behaviour shown in screenshot 2.

Screenshots
Screenshot 1:

Screenshot 2:

Environment (please complete the following information):

• Device: Intel cloud server ncl-edu
• OS: Ubuntu
• Lava version lava_loihi-0.7.0

Further context
I have a fix for this bug and will make a PR right away.