Reduce malloc of iterators in Get() code paths

Summary:
This patch optimized Get() code paths by avoiding malloc of iterators. Iterator creation is moved to mem table rep implementations, where a callback is called when any key is found. This is the same practice as what we do in (SST) table readers.

db_bench result for readrandom following a writeseq, with no compression, single thread and tmpfs, we see throughput improved to 144958 from 139027, about 3%.

Test Plan: make all check

Reviewers: dhruba, haobo, igor

Reviewed By: haobo

CC: leveldb, yhchiang

Differential Revision: https://reviews.facebook.net/D14685

include/rocksdb/db.h

@@ -31,6 +31,7 @@ struct ReadOptions;
 struct WriteOptions;
 struct FlushOptions;
 class WriteBatch;
+class Env;
 
 // Metadata associated with each SST file.
 struct LiveFileMetaData {

include/rocksdb/memtablerep.h

@@ -41,6 +41,7 @@
 namespace rocksdb {
 
 class Arena;
+class LookupKey;
 class Slice;
 class SliceTransform;
 
@@ -74,6 +75,20 @@ class MemTableRep {
   // nothing.
   virtual void MarkReadOnly() { }
 
+  // Look up key from the mem table, since the first key in the mem table whose
+  // user_key matches the one given k, call the function callback_func(), with
+  // callback_args directly forwarded as the first parameter, and the mem table
+  // key as the second parameter. If the return value is false, then terminates.
+  // Otherwise, go through the next key.
+  // It's safe for Get() to terminate after having finished all the potential
+  // key for the k.user_key(), or not.
+  //
+  // Default:
+  // Get() function with a default value of dynamically construct an iterator,
+  // seek and call the call back function.
+  virtual void Get(const LookupKey& k, void* callback_args,
+                   bool (*callback_func)(void* arg, const char* entry));
+
   // Report an approximation of how much memory has been used other than memory
   // that was allocated through the arena.
   virtual size_t ApproximateMemoryUsage() = 0;