Reconsider storing code tree structure for tests by default

[IndrajeetPatil]

That is, consider whether setting write_tree = FALSE in test_collection(), and maybe turn it on to TRUE only in interactive contexts where these structures might help debug some issue?

The current approach tends to produce large git diffs for even minor changes in test code.

Is there any downside to doing so?

[lorenzwalthert]

Taking a closer look at the diff, many seem unrelated to your changes, and more related to some changes in {data.tree} or the parser itself.

I have hardly ever used these tree files in practice. And they are also .Rbuildignored. I think people can generate them if they need, as you said, so we could .gitignore them and make the default write_tree = FALSE (currently NA, which accounted for the fact that {data.tree} is only available for R >= 3.2, which is irrelevant nowadays).

[lorenzwalthert]

Once this is done, we can rebase #1195 on main to avoid its large tree diff.