Further performance improvements

Further removed multiple pointer instances
Re-use decompression readers (except zstd due to bugs)

internal/decompress/lz4.go

@@ -3,13 +3,28 @@ package decompress
 import (
 	"bytes"
 	"io"
+	"sync"
 
 	"github.com/pierrec/lz4/v4"
 )
 
-type Lz4 struct{}
+type Lz4 struct {
+	pool sync.Pool
+}
 
-func (l Lz4) Decompress(data []byte) ([]byte, error) {
-	rdr := lz4.NewReader(bytes.NewReader(data))
+func NewLz4() *Lz4 {
+	return &Lz4{
+		pool: sync.Pool{
+			New: func() any {
+				return lz4.NewReader(nil)
+			},
+		},
+	}
+}
+
+func (l *Lz4) Decompress(data []byte) ([]byte, error) {
+	rdr := l.pool.Get().(*lz4.Reader)
+	defer l.pool.Put(rdr)
+	rdr.Reset(bytes.NewReader(data))
 	return io.ReadAll(rdr)
 }