fix: writer/reader correctness and determinism issues

- merge: cloneTable keeps relationships; skip-tables applies to new schemas
- diff: detect schema description/owner changes
- prisma: reader no longer turns relation fields into columns, enum
  detection via declared names; writer type mapping is ordered
- typeorm writer: keep explicit SQL types that cannot be inferred
- drizzle: enum columns call the enum constant; reader resolves them
- mysql/mssql/sqlite writers: honour OutputPath, deterministic column
  and constraint order; mssql live execute covers full schema
- template: ToYAML recovers from panics; Merge nil-pointer loop
- regenerate drizzle fixtures

pkg/readers/prisma/reader.go

@@ -13,7 +13,8 @@ import (
 
 // Reader implements the readers.Reader interface for Prisma schema format
 type Reader struct {
-	options *readers.ReaderOptions
+	options   *readers.ReaderOptions
+	enumNames map[string]bool // enum names declared in the schema being parsed
 }
 
 // NewReader creates a new Prisma reader with the given options
@@ -82,6 +83,8 @@ func (r *Reader) parsePrisma(content string) (*models.Database, error) {
 	schema := models.InitSchema("public")
 	schema.Enums = make([]*models.Enum, 0)
 
+	r.enumNames = collectEnumNames(content)
+
 	scanner := bufio.NewScanner(strings.NewReader(content))
 
 	// State tracking
@@ -600,20 +603,21 @@ func (r *Reader) isPrimitiveType(typeName string) bool {
 	return false
 }
 
-// isEnumType checks if a type name might be an enum
-// Note: We can't definitively check against schema.Enums at parse time
-// because enums might be defined after the model, so we just check
-// if it starts with uppercase (Prisma convention for enums)
-func (r *Reader) isEnumType(typeName string, table *models.Table) bool {
-	// Simple heuristic: enum types start with uppercase letter
-	// and are not known model names (though we can't check that yet)
-	if len(typeName) > 0 && typeName[0] >= 'A' && typeName[0] <= 'Z' {
-		// Additional check: primitive types are already handled above
-		// So if it's uppercase and not primitive, it's likely an enum or model
-		// We'll assume it's an enum if it's a single word
-		return !strings.Contains(typeName, "_")
+// isEnumType reports whether typeName is an enum declared in the schema.
+// Enum names are collected up front because enums may be declared after the
+// models that use them.
+func (r *Reader) isEnumType(typeName string, _ *models.Table) bool {
+	return r.enumNames[typeName]
+}
+
+var enumDeclRegex = regexp.MustCompile(`(?m)^\s*enum\s+(\w+)\s*{`)
+
+func collectEnumNames(content string) map[string]bool {
+	names := make(map[string]bool)
+	for _, m := range enumDeclRegex.FindAllStringSubmatch(content, -1) {
+		names[m[1]] = true
 	}
-	return false
+	return names
 }
 
 // createConstraintFromRelation creates a FK constraint from a @relation attribute