Performance Measurments Slow

I recently ran a comparison between different coin-selection implementations, namely bitcoin-core, rust-bitcoin-coin-selection and this crate, bdk coin select. In so doing, I noticed BDK coin-select is slow, like really really slow.

These are the numbers I recently recorded:
| implementation | interface | pool size|ns/select |
|------------------------|---------------------|----------|----------|
| bitcoin-coin-selection | select_coins() | 400 |862,910 |
| bitcoin core | AttemptSelection() | 400 |16,695,408|
| bdk coin-select | run_bnb(lowest_fee) | 50 |30,202,000|

To make matters worse, both bitcoin-core and bitcoin-coin-selection actually run the interface in a loop [10 times](https://github.com/bitcoin/bitcoin/blob/59224b66aa1db43adc61c15ee41b413951f22f80/src/bench/coin_selection.cpp#L133) for each measurement.

For more details on the measurements, see my PR for adding coin-selection Criterion measurements [here](https://github.com/p2pderivatives/rust-bitcoin-coin-selection/pull/320).

Hopefully these findings can help you all in fine tuning the performance of this crate.

Start by reproducing the Criterion measurements described in the linked rust-bitcoin-coin-selection PR, focusing on the bdk coin-select run_bnb(lowest_fee) case and the reported pool sizes. Compare its results with the other implementations and identify the source of the slowdown; done means the measurements improve and the benchmark results document the change.

Written by the indexing model from the issue text.