Use goreleaser to build and release (#244)

Instead of using gox on one side and an action to release, we can merge
them together with goreleaser which will build for extra targets (arm,
mips if needed in the future) and it also takes care of creating
checksums, a source archive, and a changelog and creating a release with
all the artifacts.

All binaries should respect the old naming convention, so any scripts
out there should still work.

Signed-off-by: Itxaka <igarcia@suse.com>

vendor/github.com/klauspost/compress/zstd/bitreader.go

@@ -5,6 +5,7 @@
 package zstd
 
 import (
+	"encoding/binary"
 	"errors"
 	"io"
 	"math/bits"
@@ -34,8 +35,12 @@ func (b *bitReader) init(in []byte) error {
 	}
 	b.bitsRead = 64
 	b.value = 0
-	b.fill()
-	b.fill()
+	if len(in) >= 8 {
+		b.fillFastStart()
+	} else {
+		b.fill()
+		b.fill()
+	}
 	b.bitsRead += 8 - uint8(highBits(uint32(v)))
 	return nil
 }
@@ -63,21 +68,31 @@ func (b *bitReader) fillFast() {
 	if b.bitsRead < 32 {
 		return
 	}
-	// Do single re-slice to avoid bounds checks.
-	v := b.in[b.off-4 : b.off]
+	// 2 bounds checks.
+	v := b.in[b.off-4:]
+	v = v[:4]
 	low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24)
 	b.value = (b.value << 32) | uint64(low)
 	b.bitsRead -= 32
 	b.off -= 4
 }
 
+// fillFastStart() assumes the bitreader is empty and there is at least 8 bytes to read.
+func (b *bitReader) fillFastStart() {
+	// Do single re-slice to avoid bounds checks.
+	b.value = binary.LittleEndian.Uint64(b.in[b.off-8:])
+	b.bitsRead = 0
+	b.off -= 8
+}
+
 // fill() will make sure at least 32 bits are available.
 func (b *bitReader) fill() {
 	if b.bitsRead < 32 {
 		return
 	}
 	if b.off >= 4 {
-		v := b.in[b.off-4 : b.off]
+		v := b.in[b.off-4:]
+		v = v[:4]
 		low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24)
 		b.value = (b.value << 32) | uint64(low)
 		b.bitsRead -= 32