feat: add --types flag and stdlib nullable type support for bun/gorm writers

* Fix pgsql reader double-quoting defaults: normalizePostgresDefault strips
  surrounding SQL string literal quotes from column_default before storing,
  matching the convention used by every other reader.

* Add NullableTypes field to WriterOptions with NullableTypeResolveSpec
  (default) and NullableTypeStdlib constants.

* Both bun and gorm TypeMappers now accept a typeStyle parameter. stdlib
  mode produces sql.NullString/NullInt32/NullTime etc. for nullable scalars,
  plain Go slices for arrays, and time.Time for NOT NULL timestamps. Default
  resolvespec behaviour is unchanged.

* Add --types flag to convert and split commands.

* Update bun/README.md and gorm/README.md with side-by-side generated code
  examples, updated type mapping tables, and Writer Options documentation.

cmd/relspec/split.go

@@ -22,6 +22,7 @@ var (
 	splitDatabaseName  string
 	splitExcludeSchema string
 	splitExcludeTables string
+	splitNullableTypes string
 )
 
 var splitCmd = &cobra.Command{
@@ -110,6 +111,7 @@ func init() {
 	splitCmd.Flags().StringVar(&splitTables, "tables", "", "Comma-separated list of table names to include (case-insensitive)")
 	splitCmd.Flags().StringVar(&splitExcludeSchema, "exclude-schema", "", "Comma-separated list of schema names to exclude")
 	splitCmd.Flags().StringVar(&splitExcludeTables, "exclude-tables", "", "Comma-separated list of table names to exclude (case-insensitive)")
+	splitCmd.Flags().StringVar(&splitNullableTypes, "types", "", "Nullable type package for code-gen writers (bun/gorm): 'resolvespec' (default) or 'stdlib' (database/sql)")
 
 	err := splitCmd.MarkFlagRequired("from")
 	if err != nil {
@@ -185,6 +187,7 @@ func runSplit(cmd *cobra.Command, args []string) error {
 		splitPackageName,
 		"",    // no schema filter for split
 		false, // no flatten-schema for split
+		splitNullableTypes,
 	)
 	if err != nil {
 		return fmt.Errorf("failed to write output: %w", err)