optimize validate_removals()

[arvidn]

Purpose:

transition validate_removals() to use the fast function to compute a merkle tree root. The MerkleSet class is an inefficient implementation, and very expensive to use. The block validation as well as validate_additions() were transitioned to use the fast function when I first introduced it, but must have missed this place.

history

in #10146 Mariano introduced an assert removals_res.proofs is not None which looks wrong, because the function that the proof is passed to handles the case where it's None. That's specifically the case this PR focuses on updating.

The fact that nothing has been broken this whole time (as far as I can tell) may suggest that perhaps the feature of omitting the proof when all removals are returned, is unnecessary.

This patch operates under the assumption that it is useful. Improves it and removes the assert blocking it.

Current Behavior:

validate_removals() is expensive.
The response to a removals request must include proofs, even when every single removal is included.

New Behavior:

validate_removals() is less expensive.
The response to a removals request may omit proofs if every single removal is included.

Testing Notes:

Tests have been added.