Migrate remaining data structs to fully borrowed

[sffc]

Some data structs are capable of zero-copy deserialization. We should migrate all data structs to be this way. This means removing #[yoke(cloning_zcf)] where possible (from #847).

Separate issues to track each component:

• Write tooling to keep track of (and enforce) that our data structs do not allocate #1661
• DateTimeFormat (see Migrate all DateTimeFormat data structs to zero-copy #876)
• PluralRules (see PluralRuleStringsV1 should be parsed at data load time #615)
• UnicodeSet (see Support zero-copy in UnicodeSet #855)
• LocaleCanonicalizer (see Migrate LocaleCanonicalizer data structs to zero-copy #1034)
• Calendar (see Convert Japanese calendar data to zero-copy #1393)
• LSTM (see Migrate LSTM segmenter to zero-copy #1426)

Before closing this issue, the code should be audited to ensure there are no remaining instances of #[yoke(cloning_zcf)], and then this meta-issue can be closed.

[Manishearth]

For some structs this will be dependent on #844

[sffc]

ICU4X 0.4 is shipped. This got done in some crates but not others. Punting to 0.5.

[robertbastian]

In addition to the issues above there are also #[yoke(cloning_zcf)]s on UnicodeSet (fixing in #1604) and in experimental/segmenter_lstm. Should we file an issue for the latter?

[Manishearth]

We can also just add it as a list entry there without filing an issue, either works

[sffc]

Segmenter LSTM data is tracked by #905. That issue is assigned to Ting-yu and scheduled for Sprint C. I added it to the list in the OP.

[robertbastian]

Should be #1426 actually

[Manishearth]

#1676

[Manishearth]

Here's the current list of non-zc keys:

    "datetime/skeletons@1",
    "list/and@1",
    "list/or@1",
    "list/unit@1",
    "locale_canonicalizer/aliases@1",
    "locale_canonicalizer/likelysubtags@1",

The first one is rather tricky, I plan to file an issue about it (#1678). The next three are experimental, the last two are #1034 and need LocaleStr (#831)

[Manishearth]

list_formatter: #1680

[sffc]

All data structs are now zero-copy except for skeletons. LSTM is not yet plugged into the data provider. I'm going to close this issue as completed, since the objective is complete, but ongoing work can be tracked in:

• Migrate all DateTimeFormat data structs to zero-copy #876
• Migrate LSTM segmenter to zero-copy #1426

[robertbastian]

Actually #1678 is the leaf, I think we can close #876 as well, we're not tracking anything there either.

[sffc]

#876 should stay open until the skeletons / components bag is actually zero-copy (no ZeroFrom hacks).