Add child node performance

[heckendorfc]

I generate XML documents which can easily have nodes with >10k children. With the current logic in this package, there is no way to add child nodes without iterating through or creating R objects containing the complete list of children.

I restructured xml_add_child so that using .where=0L (prepend mode) will bypass these very slow function calls.

With the existing code, using .where=0L will produce an error if there are no children, rather than just insert the node as the first and only child. I assume this is a bug so I also included a fix for this in my changes.

[jimhester]

This looks good, thank you for the PR! Could you add a test for the .where = 0 case with no children to tests/testthat/test-modify-xml.R? Also please add a note to NEWS.md referencing this PR # and your GitHub handle. Thanks!

[heckendorfc]

I can't reproduce that error I was talking about, so I'm going to guess I'm either thinking about something in XML rather than this package, or something I encountered during development. It's been a few days since I worked on this code.

The .where = 0L case should already be handled by the first test in:
"xml_add_child can insert anywhere in the child list"

Recent commit adds NOTES message and fixes the improper text node handling the tests picked up.

edit: oops, NEWS, not NOTES

[jimhester]

Thanks!