pgio: optimize append/set functions with direct byte shifts

Replace zero-fill + overwrite pattern with direct byte appends.
Use array pointer cast in SetInt32 to eliminate bounds checks.

Benchmarks show 12-19% improvement in message building:
  BindMessage5Params:  164 ns → 132 ns  (19% faster)
  BindMessage20Params: 922 ns → 780 ns  (15% faster)
  Simulate10kQueries: 1.37ms → 1.21ms  (12% faster)

Assembly comparison for SetInt32:
  Before: 134 bytes, 4 bounds checks, 4 panicIndex calls
  After:   63 bytes, 1 bounds check,  1 panicSliceConvert call

No functional changes. All existing tests pass.

Author: analytically

internal/pgio/write.go

@@ -1,26 +1,18 @@
 package pgio
 
-import "encoding/binary"
-
 func AppendUint16(buf []byte, n uint16) []byte {
-	wp := len(buf)
-	buf = append(buf, 0, 0)
-	binary.BigEndian.PutUint16(buf[wp:], n)
-	return buf
+	return append(buf, byte(n>>8), byte(n))
 }
 
 func AppendUint32(buf []byte, n uint32) []byte {
-	wp := len(buf)
-	buf = append(buf, 0, 0, 0, 0)
-	binary.BigEndian.PutUint32(buf[wp:], n)
-	return buf
+	return append(buf, byte(n>>24), byte(n>>16), byte(n>>8), byte(n))
 }
 
 func AppendUint64(buf []byte, n uint64) []byte {
-	wp := len(buf)
-	buf = append(buf, 0, 0, 0, 0, 0, 0, 0, 0)
-	binary.BigEndian.PutUint64(buf[wp:], n)
-	return buf
+	return append(buf,
+		byte(n>>56), byte(n>>48), byte(n>>40), byte(n>>32),
+		byte(n>>24), byte(n>>16), byte(n>>8), byte(n),
+	)
 }
 
 func AppendInt16(buf []byte, n int16) []byte {
@@ -36,5 +28,5 @@ func AppendInt64(buf []byte, n int64) []byte {
 }
 
 func SetInt32(buf []byte, n int32) {
-	binary.BigEndian.PutUint32(buf, uint32(n))
+	*(*[4]byte)(buf) = [4]byte{byte(n >> 24), byte(n >> 16), byte(n >> 8), byte(n)}
 }