couchdb beam process doesn't start dir_size, gives very poor performance when scaling buckets up.

Issue found when running with 60 buckets and profiling all CPU intensive processes.

fprof showed the per bucket couchdb_stats_reader code was spending a lot of time doing "dir_size:get", and with > 10 buckets this cost grows into a very CPU hungry beam. Note 10 buckets and things aren't too bad.

Some analysis showed we were going down the dir_size:get_slow path because the dir_size module was not running in the couchdb beam process.

53 get(Dir) ->
54 case erlang:whereis(?MODULE) of <--- returning undefined
55 undefined ->
56 get_slow(Dir); <- yes this is SLOW!
57 _Pid ->
58 gen_server:call(?MODULE,

, infinity)
59 end.