Write an XPath for extended known-literal expressions

[MichaelChirico]

Follow-up to #957 (comment)

There are a few places that would benefit from some shared logic about what constitutes a "literal expression" along the lines of these:

const_exprs <- c(
  '"a"',
  '1',
  '1+1i',
  'c("a")',
  'c("a", 1)',
  'c("a", 1+1i)',
  'rep("a", 2)',
  'rep("a", n)',
  'rep(c("a", "b"), 5)',
  'c(rep("a", 10), rep("b", 10))'
)

but excluding non-constant expressions like:

non_constant_exprs <- c(
  'a',
  'a + 1',
  'foo(1)',
  'foo(c(1, 2))'
)

I think c() and rep() are the most common ways to build constant expressions, so we can limit it to combinations of those. Note that I consider rep(1:3, n) as being "constant" here, as the variable part is "constant enough" for the places where we'd use this:

• strings_as_factors_linter which looks for data.frames declared with known-character columns
• expect_{type,length,named,s3_class}_linter which look for literal exprs in the 1st argument
• yoda_test_linter which does the same

[AshesITR]

I'm not sure whether it's good to call the rep(1:3, n) expression a constant (but willing to call it "almost constant").
Regarding functions used: 1:3 == seq_len(3) is what I default to due to 1:n being bad for n == 0.
This means we could whitelist seq, seq_len, seq_along and seq.int as well.