Process edge attributes from networks graph based on user kwargs

#### Is your feature request related to a problem? Please describe.
This is my use case - I have a geodataframe with a bunch of linestring geometry. While I can visualize that with geodataframe.hvplot, I do want to show the network topology (arrows etc), so felt ploting a networkx directed graph might be better. So I used momepy.gdf_to_nx. This either returns a DiGraph or a MultiDiGraph, and converts all columns of the gdf to attributes of the graph. For my purposes a DiGraph seems sufficient. I also want to color the edges and show some attributes on hover (thanks to all the posts on ispection policy and hovertools I am able to achieve this). Since there is a geometry attribute in my initial gdf, I get an error when I visualize this graph - TypeError: '<' not supported between instances of 'LineString' and 'LineString'. [Entire stack trace is here](https://pastebin.com/Fj1aKPyF).

#### Describe the solution you'd like
I think there are a few ways to get around this, all with some changes in the [from_networkx function](https://github.com/holoviz/hvplot/blob/47f89eed102acd5b55094ff6e691672664b68aef/hvplot/networkx.py#L49):

1. Ensure we only parse through types which can be compared (and not avoid just list/dict).
2. Let the user specify edge attributes (as a list and passed in as a kwarg) from the graph to parse into the holoview Graph object.

#### Additional context
Initially I was using the defaults in momepy.gdf_to_nx and it returns a MultiDiGraph, which stores attributes one level deeper, so supporting that would also involve the keyword approach along with checking the type of graph.

There is a from_networkx function in hvplot.networkx as well as in holoviews graph.py which both seem replicas of each other, so we may have to make changes in the best possible location.