Add insta-based snapshot tests that cover side-by-side format

[yuja]

I'm going to fix inline diff issue #704, and found that the inline format isn't covered by CLI tests. I could duplicate compare_all.sh to test various formats, but it's uneasy to review that a new formatting is better than the previous version. I think snapshot testing will help guarantee the output quality without increasing maintenance burden too much.

If we like the idea, maybe we can migrate some of compare_all.sh tests to the insta-based ones.

Note that this will add ~10sec to "cargo test" runs.