Impl Serde for PackedPatternsV1 with packed-like human form

[sffc]

Replaces #5592
Depends on #5621
Depends on #5622

This PR does not export impl Deserialize for PluralElements, which was going to prevent #5592 from landing. However, if we ever add plural packed patterns, we're going to need something like that.

(note on my coding velocity: this PR took me just over 3 hours to write, which I consider to be a lot of time for a serde impl. the previous PR took less than an hour.)

[sffc]

Review note: the last commit, 674c0b4, re-writes a lot of what I wrote in d679710. You may want to review both commits at the same time.

[sffc]

I'll add @robertbastian as an optional reviewer since you were reviewer on the previous PR, or you can defer to @Manishearth.

[robertbastian]

Is there any context for this? I don't understand why you are spending 3h on a pretty serde impl on an unused type that does not affect testdata.

[dpulls]

🎉 All dependencies have been resolved !

[sffc]

Is there any context for this? I don't understand why you are spending 3h on a pretty serde impl on an unused type that does not affect testdata.

It will replace the thing currently called PackedSkeletonDataV1 but I can't switch it out until all the impls are landed.

[sffc]

I spent 3h on a pretty serde impl because my first attempt added API surface we didn't want to add, so I had to write a manual serde impl. It has been important for the datetime patterns to be human-readable in JSON, and they wouldn't have been human-readable if I had used the default human-readable serialization of the machine struct.

[robertbastian]

It will replace the thing currently called PackedSkeletonDataV1 but I can't switch it out until all the impls are landed.

Can you link an issue?

[sffc]

It will replace the thing currently called PackedSkeletonDataV1 but I can't switch it out until all the impls are landed.

Can you link an issue?

It's done now. #5648