Support multiple inputs

The current implementation assumes that there always exists a huge input that we want and can use to queue transactions on top of each other.

We need to teach the faucet to consume other inputs to minimize the dust.

No files or tests are named. Start by locating the faucet's current input-selection and transaction-queue entry points, then trace how it chooses the large input. Done means the faucet can consume multiple inputs when needed and reduces dust without breaking transaction queuing; add or update focused tests if the existing suite covers this behavior.

Written by the indexing model from the issue text.