chore: fix poor performance and long compile times in value_note.dere…

…ment() (#6523)

#6405 added a very
complicated call (`get_npk_m_hash`) to `destroy_note`, not realizing
that `destroy_note` is called in a loop by `decrement`. The loop
unrolling caused multiple calls to this function, even though they all
had the same arguments, ultimately resulting in humongous RAM usage
during compilation (over 40GB just for this contract) and very
compilation times (from 30s to multiple minutes).

This PR fixes this simply inlining the code for `destroy_note` and
fetching the key once at the beginning. It does worry me slightly
however that such a large performance hit was not noticed - this likely
affected test times as well.

noir-projects/aztec-nr/value-note/src/utils.nr

@@ -52,7 +52,18 @@ pub fn decrement_by_at_most(
     let mut decremented = 0;
     for i in 0..opt_notes.len() {
         if opt_notes[i].is_some() {
-            decremented += destroy_note(balance, owner, opt_notes[i].unwrap_unchecked());
+            let note = opt_notes[i].unwrap_unchecked();
+
+            // This is similar to destroy_note, except we only compute the owner_npk_m_hash once instead of doing it in
+            // each loop iteration.
+
+            // Ensure the note is actually owned by the owner (to prevent user from generating a valid proof while
+            // spending someone else's notes).
+            // TODO (#6312): This will break with key rotation. Fix this. Will not be able to pass this after rotating keys.
+            assert(note.npk_m_hash.eq(owner_npk_m_hash));
+            decremented += note.value;
+
+            balance.remove(note);
         }
     }