Fix comments rewritten too long

[matthewhughes934]

• Simplify test case for issue 5023

  By replacing the code quoted from the issue with a single comment that
  more directly exposes the issue. This is to

  1. Make the underlying issue easier to reason about
  2. Make this test not behave upon the behaviour of indentation length
     when rewriting lines (my main motivation for this change, since it
     makes the followup patch simpler)

  This was tested by partially reverting
  368a9b7 with the patch:

  diff --git i/src/string.rs w/src/string.rs
  index 3b971188..561e70ac 100644
  --- i/src/string.rs
  +++ w/src/string.rs
  @@ -278,9 +278,6 @@ fn break_string(max_width: usize, trim_end: bool, line_end: &str, input: &[&str]
           }
           cur_index
       };
  -    if max_width_index_in_input == 0 {
  -        return SnippetState::EndOfInput(input.concat());
  -    }
  
       // Find the position in input for breaking the string
       if line_end.is_empty()
  

  The re-running tests, or executing rustmft directly on this test,
  triggered the panic from the original bug.

• Fix wrapping of comments sometimes making too long lines

  In the case where comment_width is less than max_width and we want
  to wrap right on comment_width: rustfmt would consider it's wrapping
  length to be exactly comment_width, ignoring the length of a trailing
  indent. The fix is essentially how similar formatting code behaves, see
  e.g. shape::Shape::comment.

  This involved rewriting some existing tests, without impacting what
  they're meant to be asserting on. This emphasises that this patch is
  not backwards compatible: while I believe the change to be correct
  (some tests had comments longer than comment_width that weren't being
  formatted). So there's a trade-off to be made between correctness and
  stability, since comment_width is still not stabilised this may be
  acceptable.

  Issue: wrap_comments partially ignores comment_width with bullet lists #6801