remove_channel_in_nodes (and channel_failed with !is_permanent) don't update lowest_inbound_channel_fees

[TheBlueMatt]

While both of these shouldn't cause incorrect routing behavior, they may result in the lowest_inboud_channel_fees value being spuriously high, which could cause slower than optimal routing.

I don't think the right solution here is to just fix the behavior in question, though, I think we really should revisit the lowest_inbound_channel_fees behavior entirely - when I last benchmarked the difference between A* and dijkstra's performance on our router A* was a good bit faster, but our router has changed a lot since then, and I think there's a good chance A* isn't faster anymore (it does imply an extra map lookup during routing!). Before we yank it out entirely, though, I think we should look into storing the lowest_inbound_channel_fees field in the per-channel data as well, avoiding the extra map lookup, which may leave A* a lot faster still.

[naumenkogs]

How exactly having lowest_inbound_channel_fees on the per-channel basis would avoid network_nodes.get(&$src_node_id).map(|node| node.lowest_inbound_channel_fees), unless you rework a big chunk of the algorithm?

I don't mind algo rework, but your mention of it sounds more natural and easy than what I see in the code, so just double-checking we're talking about the same thing.

[TheBlueMatt]

In the common case (ie calling add_entries_to_cheapest_to_target_node), we're already looking at network_channels and have looked up the channel in the channels map. In that case, we could take the min-fees from the channel map and pass it through to add_entry, we'd just have to have other add_entry calls pass a None or so.