Optimization: use RBTree's iterator.update in put and batch

[vweevers]

The iterators of functional-red-black-tree have an update(value) method. For keys that already exist, memdown can use this method to directly update the corresponding node, instead of removing it (for which RBTree itself uses an iterator) and then inserting a new node.

Performance of inserting new keys is the same, overwriting keys is 1.5 to twice as fast.

A benchmark ("remove and insert" is the current implementation, "update or insert" the new):

> matcha put-bench.js
                      put new key                  
              32 op/s » remove and insert          
              33 op/s » update or insert           

                      put existing key             
             270 op/s » remove and insert          
             486 op/s » update or insert           

> matcha batch-bench.js
                      put new key
              36 op/s » remove and insert
              36 op/s » update or insert

                      put new key and delete
              27 op/s » remove and insert
              27 op/s » update or insert

                      delete non-existing keys
             152 op/s » remove and insert
             166 op/s » update or insert

                      put existing key
             342 op/s » remove and insert
             590 op/s » update or insert

[nolanlawson]

Build is failing because Saucelabs only runs for contributors... it's a dumb limitation, and I'll get around it by just opening a PR myself with your exact commits. Once that's green, I'm +1.

[nolanlawson]

https://github.com/Level/memdown/pull/45

[nolanlawson]

45 is green, so this is good to go. Thanks! :)