zstd: Make load(32|64)32 safer and smaller (#788)

load3232 and load6432 now limit their slice to its length. This improves
safety, since encoders can no longer look at the part beyond the length.
It also produces several dozen fewer instructions for bounds checks.

zstd/zstd.go

@@ -128,11 +128,11 @@ func matchLen(a, b []byte) (n int) {
 }
 
 func load3232(b []byte, i int32) uint32 {
-	return binary.LittleEndian.Uint32(b[i:])
+	return binary.LittleEndian.Uint32(b[:len(b):len(b)][i:])
 }
 
 func load6432(b []byte, i int32) uint64 {
-	return binary.LittleEndian.Uint64(b[i:])
+	return binary.LittleEndian.Uint64(b[:len(b):len(b)][i:])
 }
 
 type byter interface {