draft: Prevent lots of empty inline snapshots with --force-update-snapshots without --accept

[max-sixty]

I realize what you mean in the prior comment. Currently running with --accept is kinda painful. I'm not sure how common that is, but would be nice to fix.

Had a go for 30 mins, need to run, but pushing an initial effort and a pointer to the thing that needs fixing

[max-sixty]

OK, I think I understand the underlying problem here. In a test like:

assert_snapshot!("test", @r#test#");

• The macro invocation only sees the test value; it doesn't know about the r#...#
• insta can't find out about the r#...#, only cargo-insta, because only cargo-insta has the rust parser
• We need to compare the full r#...# to correctly decide whether a snapshot requires updating when --force-update-snapshots is passed, and we need to do that with insta only
• Unless we can run matches_fully correctly, we default to creating pending snapshots for snapshots which don't need updating, which is what the existing master does

A couple of ideas, none are great:

• Have --force-update-snapshots accept by default. This is reasonable UX — it's generally something folks are not running on every loop, but rather an occasional clean-up.
  • Currently, running --force-update-snapshots --accept is a reasonable experience — it prints a few too many names of snapshots it's updating, but that's the only cost.
  • It also sounds like that's what it's doing already, or we could pick a new name --force-accept if we wanted
• Revert Enable inline snapshots to be force-updated #569, give up on force-updating inline snapshots.
• Offer some additional option; --force-update-inline-snapshots. This would be a shame.

My current thinking is to revert #569, and then default to --force-update-snapshots accepting over a couple of versions.

[max-sixty]

I think this is broadly fixed in #581