merger: don't create key_def/merger for one source

A merge source has the same API as merger itself, so there is no
difference in behaviour between a merger created from one source and
this source itself. However there is no overhead for creating key_def,
merger and passing tuples over the merger.

This optimization gives me 13% boost on the case from #220.

Follows up #220

CHANGELOG.md

@@ -15,6 +15,9 @@ and this project adheres to [Semantic Versioning](http://semver.org/spec/v2.0.0.
   testing. It means that methods `crud.min()` and `crud.max` won't work for old
   versions (< 1.10.8) anymore.
 * Names of errors generated by CRUD operations have been unified.
+* Opmimize `crud.select()` / `crud.pairs()` for one replicaset case (say, when
+  `bucket_id` is passed or deducible from conditions). It gives 13% boost on
+  the case from #220.
 
 ### Fixed
 

crud/select/merger.lua

@@ -163,6 +163,14 @@ local function new(replicasets, space, index_id, func_name, func_args, opts)
         table.insert(merger_sources, source)
     end
 
+    -- Trick for performance.
+    --
+    -- No need to create merger, key_def and pass tuples over the
+    -- merger, when we have only one tuple source.
+    if #merger_sources == 1 then
+        return merger_sources[1]
+    end
+
     local keydef = Keydef.new(space, opts.field_names, index_id)
     -- When built-in merger is used with external keydef, `merger_lib.new(keydef)`
     -- fails. It's simply fixed by casting `keydef` to 'struct key_def&'.