runtime: faster allocator, garbage collector

GC is still single-threaded.
Multiple threads will happen in another CL.

Garbage collection pauses are typically
about half as long as they were before this CL.

R=brainman, iant, r
CC=golang-dev
https://golang.org/cl/3975046

src/pkg/runtime/debug.go

@@ -69,7 +69,8 @@ type MemStatsType struct {
 
 	// Per-size allocation statistics.
 	// Not locked during update; approximate.
-	BySize [67]struct {
+	// 61 is NumSizeClasses in the C code.
+	BySize [61]struct {
 		Size    uint32
 		Mallocs uint64
 		Frees   uint64