Extend boolean_arithmetic_linter to detect all()-like cases too

[MichaelChirico]

Follow-up to #1579

Extend boolean_arithmetic_linter() to include all()-alike expressions like sum(x == y) == length(x) --> all(x == y).

The issues are

1. we can't just test for length() on the RHS (e.g. sum(x == y) == length(z) could be totally different)
2. looking for sum(x == y) == length(x == y) (i.e., exact expression matching, which we do in some other places) will only find a particularly silly manifestation of the issue
3. there is a bit of a combinatorial explosion if we have to look for any symbol found inside the logical expression on the RHS (e.g. sum(x == y) == [length(x) _or_ length(y)]).

A first pass would be to check how many hits there are for the naive version (i.e., condition 1), to see how worth investing in this complicated logic is to begin with.

[MichaelChirico]

Adding potential test cases:

test_that("boolean_arithmetic_linter blocks comparisons to the object length", {
  expect_lint(
    "length(which(condition)) < length(condition)",
    "grep\\(pattern, x\\) is better than which\\(grepl\\(pattern, x\\)\\)\\.",
    boolean_arithmetic_linter
  )
  expect_lint(
    "length(which(is_treatment)) == length(is_treatment)",
    "grep\\(pattern, x\\) is better than which\\(grepl\\(pattern, x\\)\\)\\.",
    boolean_arithmetic_linter
  )
  expect_lint(
    "length(grep(pattern, x)) < length(x)",
    "grep\\(pattern, x\\) is better than which\\(grepl\\(pattern, x\\)\\)\\.",
    boolean_arithmetic_linter
  )
  expect_lint(
    "sum(x == y) < length(x)",
    "grep\\(pattern, x\\) is better than which\\(grepl\\(pattern, x\\)\\)\\.",
    boolean_arithmetic_linter
  )
  expect_lint(
    "sum(grepl(pattern, x)) == length(x)",
    "grep\\(pattern, x\\) is better than which\\(grepl\\(pattern, x\\)\\)\\.",
    boolean_arithmetic_linter
  )
})