Unregistered OIDs are handled the same as unknown OIDs

This improves handling of unregistered types. In general, they should
"just work". But there are performance benefits gained and some edge
cases avoided by registering types. Updated documentation to mention
this.

https://github.com/jackc/pgx/issues/1445

pgtype/pgtype.go

@@ -1326,16 +1326,16 @@ func (m *Map) PlanEncode(oid uint32, format int16, value any) EncodePlan {
 	}
 
 	var dt *Type
-
-	if oid == 0 {
+	if dataType, ok := m.TypeForOID(oid); ok {
+		dt = dataType
+	} else {
+		// If no type for the OID was found, then either it is unknowable (e.g. the simple protocol) or it is an
+		// unregistered type. In either case try to find the type and OID that matches the value (e.g. a []byte would be
+		// registered to PostgreSQL bytea).
 		if dataType, ok := m.TypeForValue(value); ok {
 			dt = dataType
 			oid = dt.OID // Preserve assumed OID in case we are recursively called below.
 		}
-	} else {
-		if dataType, ok := m.TypeForOID(oid); ok {
-			dt = dataType
-		}
 	}
 
 	if dt != nil {