refactor(linter): Let fixer functions decide whether to apply a fix or not

[DonIsaac]

Rules that have fixers for some cases, but not others, all have some logic that looks like this:

fn run<'a>(&self, ctx: &LintContext<'a>) {
  // Check for a violation
  let diagnostic = create_diagnostic_for_rule();
  let can_fix = check_if_fixable(); // <-- this is problematic
  if can_fix {
    ctx.diagnostic_with_fix(diagnostic, |fixer| do_fix(ctx, fixer, node));
  } else {
    ctx.diagnostic(diagnostic);
  }
}

This pushes a lot of fixing logic outside of the fixer. In cases where --fix is not used, these are all wasted cycles. If we move fixability checks to inside the fixer's closure, we can prevent rules from doing work that just gets thrown away.

fn run<'a>(&self, ctx: &LintContext<'a>) {
  // Check for a violation
  let diagnostic = create_diagnostic_for_rule();
  ctx.diagnostic_with_fix(diagnostic, |fixer| {
    if !check_if_fixable(ctx, fixer, node) {
      return None;
    }
    return do_fix(ctx, fixer, node);
  });
}