Remove recursive calls in `Parser` and `print` functions

[KaruroChori]

At the moment, parsing from XML and generating XML is handled by functions making use of recursion.
In theory the parser does not fall in the class of "tail recursion", so some types of documents can fail due to buffer overflows regardless of how much memory the system has. Dynamic allocation is probably the only way forward here.

The XML serialization on the other hand is using recursion out of convenience, but as the tree has full context for each node, we don't need a stack to navigate it. Hence, it can be optimized to operate with $$O\left (1\right )$$ .

[KaruroChori]

I just realized that the stack a stack is not really needed for XML parsing, nor for the builder:

• The parser is not even meant to validate. So we don't need the stack, as the closing tags will just be recognized as closing without the need to match anything in particular (but we could still keep track of the name for the current head and still without a stack)
• The builder has no need for the stack either, it could just navigate the tree it is building.

[KaruroChori]

Thanks to the new visitor pattern, since 86566df we have print2 which is no longer using precious stack nor it requires extra allocations.
After some testing, print2 and the split print_h helpers will replace the older code.