improve xdf2 parser

Improved xdf2 Schema parser performance by 45% (4.6s to 2.6s for parsing BOB on my local machine) by:

1. Not freeing memory for already processed nodes immediately
2. Optimizing the creation of generic types

Regarding 2.

a: list[str] = list()
b = list[str]()

Both variants create an empty list and communicate the full type of the list. However, the second variant evaluates list[str] every time the line is called, whereas the first line ignores the type annotation when creating the list. The first option is therefore faster when many generic types are created.