descriptors: Improve max satisfaction weight estimates for recovery path inputs

[jp1ac4]

When estimating the max satisfaction weight of a transaction input, we use a different approach depending on whether we are considering a primary or non-primary spending path. These can give different values for the same element, e.g. a Schnorr signature for a non-primary path estimate will include a byte for the sighash suffix. For primary paths, we use rust-miniscript's planning module, and I think we should use that also for non-primary paths to ensure consistency in the estimates.

Furthermore, a non-primary estimate currently considers the max across all paths, which could be the primary path still. Ideally, there would be a way to specify which spending path we are using and to get the estimate for this path only.

The impact of the above is that the fee for a recovery transaction may be higher than necessary to achieve the target feerate.