The sweeper clusters input with similar fee rates together to save on fees. However, in such a cluster all inputs won't necessarily have the exact same fee preference, and hence we must choose a "compromise fee rate" to use when sweeping them together.

Currently we take just the average among the inputs in the cluster:

As pointed out by @cfromknecht (see #4779 (comment)) this is probably not optimal, for several reasons:

• a high pref input might get a lower fee rate than we want, in case it gets clustered together with lower pri inputs.
• it is not weight aware, meaning that the fee rate for heavy inputs will impact the final fee rate no more than less heavy inputs
• the input might be lingering for a while, so the fee estimates might not be up-to-date. Could make sense to update them/or take most recent fee estimate.