Speed up the innermost btree parsing function. When extracting a key, instead of parsing into a list, and then casting that into a tuple, just pre-allocate the tuple (we know the size anyway), and then parse. This saves approx 10% of the time to parse a btree index.