implicit_assignment_linter misses non-call assignments

[MichaelChirico]

#2016 is not quite as relevant at the moment since assignments like this are always skipped:

lint("A && (B <- ...)\n", implicit_assignment_linter())

[MichaelChirico]

There are two existing tests that would be affected by the simplest fix:

lintr/tests/testthat/test-implicit_assignment_linter.R

Line 84 in 8ac582c

.f = ~ (x <- .x + 1)

lintr/tests/testthat/test-implicit_assignment_linter.R

Line 295 in 8ac582c

expect_lint("mean(x <- (y <- 1:3) + 1L)", lint_message, linter)

The latter looks like it should be two lints to me.

The former is tougher. Should we carve out an exception for {purrr} mappers? OTOH this is like a violation of function_return_linter() -- why have an assignment in a one-line function? If it is <<- (which is also covered by //LEFT_ASSIGN), it's more interesting, but we could still force { to be used instead:

x <- 1
map(
  .x = 1:4,
  .f = ~ {x <<- .x + x}
)

Of course, that's a brace_linter() violation now, so it would have to be:

x <- 1
map(
  .x = 1:4,
  .f = ~ {
    x <<- .x + x
  }
)

cc @IndrajeetPatil