Count chars when creating SourceAnnotation

[MarcusGrass]

Fixes #6083.

In annotate snippets it takes the length of the string by char count https://github.com/rust-lang/annotate-snippets-rs/blob/master/src/renderer/display_list.rs#L935, both on 0.9 and 0.10, but in the code here the
length is taken by byte-index.

Since the user's string has chars that do not correspond 1-to-1 with byte-indices there's a range-mismatch and a following panic.

It only occurs in some cases and was fairly hard to reproduce, I think it has something to do with max line-count as well, and that's when it blows up.

I'm not so familiar with this repo, I can't get the test to pass. The source can't be formatted, since the comment would be lost, but if that's disabled rustfmt will err on trailing whitespace and I don't know if that can be disabled, but the test really checks that rustfmt doesn't panic, so maybe it just doesn't fit into this mould.
E: I put the test-file under its own directory and added it to the dont_emit_ice_test instead.

Can easily be reproduced by running rustfmt tests/issue-6083/issue_6083.rs vs cargo run --bin rustfmt -- tests/issue-6083/issue_6083.rs on this branch.

[ytmimi]

@MarcusGrass thanks for looking into this! I came to the same conclusion regarding chars vs bytes when I looked into this a while ago. I'll follow up when I've had a chance to take a look at this / refresh my memory on what I was looking into previously. This has been a long standing issue and I'd be very glad if we could fix it!