Emoji alphabet (alphabet choice, not toolbar)

[willwade]

From Heide's v0.1.20 testing (Galaxy Tab S9): "I miss my emojis, though, lol! Especially in email."

Emoji is an alphabet-choice problem, like Klingon/Runic/IPA — the engine's node tree comes from the selected alphabet XML. Approaches:

1. Dedicated emoji alphabet in DasherCore data (alphabet.emoji.xml): grouped nodes (smileys, hearts, gestures, food, symbols…), uniform initial frequencies, optionally a small training corpus of common sequences. Appears in the alphabet picker like any other. Downside: switching alphabets to reach emoji mid-sentence is clunky.
2. Emoji group appended to existing alphabets (like punctuation groups): a small curated set (~100 common emoji) as an extra group in the main alphabets — reachable without switching. Needs DasherCore data changes across alphabets or a merge at load.
3. Both: appended quick set in main alphabets + full emoji alphabet for heavy users.

Blocking detail: emoji are multi-codepoint (ZWJ sequences, skin tones) — the alphabet reader must commit full sequences per node. Verify DasherCore's UTF-8 aggregation handles that (v0.2.25 hardened the reader; needs a test).

Tracking here for the Android side (alphabet picker already exposes whatever the engine registers); the data/alphabet work lands in DasherCore.

[willwade]

Decision: option 2 — toolbar toggle switching to a dedicated emoji-only alphabet.

• DasherCore#96 adds alphabet.emoji.xml (9 topical groups, ~300 nodes, ZWJ/skin-tone multi-codepoint nodes with regression tests) + training_emoji.txt
• feat(ime): emoji toggle in the editing toolbar (#61, option 2) #65 adds the 😀/Abc toggle in the editing toolbar (hidden until the data bundle ships it)

Follow-ups kept open-ended: option 3 (English with emoji trained variant — emoji frequencies learned in context) if usage warrants; option 1 (dynamic group injection API) only if switching friction shows up in real use.

[willwade]

Decision superseded by RFC 0020: emoji move into the tree for every alphabet (settings-gated load-time merge, control-mode precedent) instead of a toggle switching to a standalone alphabet. The v0.1.21 design is withdrawn: degenerate corpus mass (space = 50% of trained lookups), 31 multi-codepoint nodes permanently untrainable by the per-codepoint trainer, and an Android 13 native crash on selection. The 😀/Abc toolbar toggles (#65/#67) will be removed in the release that ships the extension. The standalone Emoji alphabet (DasherCore#96) is kept but not prioritised. Crash fix gates the release — tracked in the RFC test matrix.