Add a mode which preserves carriage returns

Here are two important properties of the `.ss` snapshot files:

- exactly character-for-character accurate, no slop around leading or trailing whitespace, no forbidden characters
- what you see in the snapshot file is exactly what you get, except for the following escaping rules which preserve the above
- the following characters are escaped: `𐝃` -> `𐝃𐝃`, `𐝁` -> `𐝃𐝁` (they are from an [untranslated dead language](https://en.wikipedia.org/wiki/Linear_A))
- if the first character on a line within a snapshot is `╔` then it is replaced with `𐝁` (this preserves ASCII art within snapshots)

This combination of character-accurate + WYSIWYG breaks down in only one place - line endings. You can't see them, and git's complex and poorly understood line-ending-mutation rules mean that most teams can't reliably do source control that differentiates between `\n` and `\r\n`.

Rather than randomly punch users in the face with this triviality, we do the following:

- Internally, every text-based snapshot in spotless-snapshot has a `.replace("\r", "")`, so snapshots will never fail because of line-ending differences
- New `.ss` files are always written using `\n` line endings
- If an `.ss` file is loaded from disk with `\r\n`, it is converted and parsed using `\n`, but will be written back to disk as `\r\n`

However, this means that the snapshots are not *exactly* character-for-character accurate because they do not preserve `\r`. For users that want to preserve `\r`, we could add a mode which preserves the `\r` character if the user uses the `.ss.asar` format

- #1

Adding support for storing `\r` within `.ss` files might be possible (maybe encode them as `𐝃r`?), but doesn't seem like a good idea.

Start by tracing how .ss files and the .ss.asar format are loaded, parsed, and written, including the existing carriage-return normalization described here. Define the mode's behavior around preserving \r and round-tripping files without changing other escaping rules; done means .ss.asar snapshots retain carriage returns while ordinary snapshots keep current behavior.

Written by the indexing model from the issue text.