Should name-unquoting work everywhere in the AST?

[lionel-]

expr(c(!! "foo" := 1))

[lionel-]

We are currently taking the approach that name-unquoting is a property of tidy dots rather than tidy eval. So it only works at top-level, e.g. list2(!!nm := foo) and exprs(!!nm := foo).

[lionel-]

We want to revisit this. One problem with this approach is that this will prevent using QQ to program on the language with data.table-like interfaces, because := will be transformed to =.

[lionel-]

We can just preserve := calls when their LHS does not use injection.