When BnB selection fails to find an exact match, we fall back to the original Bitcoin Core algorithm. This PR replaces that fallback with the Single Random Draw strategy. Additionally, because SRD uses effective values, the previous fee increasing loop thing is now removed and effective value is used everywhere.
Some tests may fail spuriously because they may rely on semi-deterministic behavior from the original coin selection algorithm. I have been able to fix the ones that fail the most, but others may still have issues. Please let me know if you see any tests fail for coin selection reasons (e.g. insufficient funds, missing inputs, etc.) and I will try to fix them.
I will be working on doing simulations this week and will post the data once I finish them.