Phase 9: ESL superpowers — Chinese gloss + tone-rewrite

Inline Chinese gloss (offline) and a "say it more naturally" / tone-rewrite,
the two ESL features for the Mandarin-speaking writer.

Gloss: embedded English→Chinese dictionary (gloss.json.gz, 57k common words
built from ECDICT via scripts/build_gloss.py). lexicon gains Gloss()/Result.Gloss
and a lightweight GET /api/gloss/{word}; the right-click WordCard leads with the
中文; GlossTip shows it on a 350ms hover (reuses wordAt, so CJK is never glossed).
Offline + instant, works with the LLM down.

Rewrite: selecting text pops a SelectionBubble (✨更自然 + the tone vocabulary);
picking a style calls POST /api/docs/:id/rewrite (llm.RunRewrite, stateless,
owner-scoped) and shows a RewritePreview (original→rewrite, accept/cancel/retry).
Accept applies it in-editor.

Tests added in lexicon and suggestions. go build/vet/test, tsc, vite all clean;
live smoke vs a fake vLLM verified gloss + rewrite + 400/404/502 paths.

Claude-Session: https://claude.ai/code/session_016Yr6jELuRc7hyzYLccQKZd

scripts/build_gloss.py

@@ -0,0 +1,106 @@
+#!/usr/bin/env python3
+"""Build the embedded English->Chinese gloss dataset from ECDICT.
+
+Petal shows an instant Chinese gloss when the writer (an English-as-a-second-
+language user whose first language is Mandarin) hovers or right-clicks a word.
+The gloss is served offline from a small map compiled into the Go binary, the
+same way the English definitions/synonyms are (see internal/lexicon).
+
+Source: ECDICT (https://github.com/skywind3000/ECDICT), MIT-licensed. We keep
+only common single words that carry a Chinese translation, trim each gloss to a
+couple of senses, and drop the noisy "[网络]" (internet-slang) lines — the result
+gzips to a few MB, in line with the other lexicon assets.
+
+Usage:
+    curl -sL https://raw.githubusercontent.com/skywind3000/ECDICT/master/ecdict.csv -o ecdict.csv
+    python3 scripts/build_gloss.py ecdict.csv internal/lexicon/data/gloss.json.gz
+"""
+import csv
+import gzip
+import json
+import re
+import sys
+
+# Frequency gate: keep a word only if ECDICT ranks it in the BNC or COCA
+# frequency lists (frq/bnc > 0). That bounds the asset to the words an ESL
+# writer actually meets, dropping the long tail of archaic/technical headwords.
+# Words below this rank but present nowhere in the frequency lists are skipped.
+MAX_RANK = 50000
+
+# A gloss is at most this many sense-lines and characters, so the hover bubble
+# stays a glance, not a wall of text.
+MAX_SENSES = 3
+MAX_CHARS = 80
+
+# Single English word: letters, with internal apostrophe/hyphen (so "don't",
+# "well-being" survive but multi-word phrases and codes are dropped).
+WORD_RE = re.compile(r"^[a-z][a-z'\-]*[a-z]$|^[a-z]$")
+
+# Tag lines we drop from a translation: "[网络]" is crowd-sourced internet slang,
+# "[俚]" slang, "[古]" archaic — none help an ESL writer pick everyday meaning.
+DROP_TAG_RE = re.compile(r"^\[(网络|俚|古|罕|废)\]")
+
+
+def clean_gloss(translation: str) -> str:
+    """Trim an ECDICT translation to a compact, glanceable Chinese gloss."""
+    # ECDICT separates senses with a literal backslash-n; normalize to real
+    # newlines (and tolerate genuine newlines) before splitting.
+    translation = translation.replace("\\n", "\n")
+    senses = []
+    for line in translation.split("\n"):
+        line = line.strip()
+        if not line or DROP_TAG_RE.match(line):
+            continue
+        senses.append(line)
+        if len(senses) >= MAX_SENSES:
+            break
+    gloss = "；".join(senses)
+    if len(gloss) > MAX_CHARS:
+        gloss = gloss[:MAX_CHARS].rstrip("；，;, ") + "…"
+    return gloss
+
+
+def rank(row: dict) -> int:
+    """Best (lowest, non-zero) frequency rank across COCA and BNC."""
+    ranks = []
+    for key in ("frq", "bnc"):
+        try:
+            v = int(row.get(key) or 0)
+        except ValueError:
+            v = 0
+        if v > 0:
+            ranks.append(v)
+    return min(ranks) if ranks else 0
+
+
+def main(src: str, dst: str) -> None:
+    out: dict[str, str] = {}
+    kept_rank: dict[str, int] = {}
+    with open(src, newline="", encoding="utf-8") as f:
+        for row in csv.DictReader(f):
+            word = (row.get("word") or "").strip().lower()
+            translation = (row.get("translation") or "").strip()
+            if not word or not translation or not WORD_RE.match(word):
+                continue
+            r = rank(row)
+            if r == 0 or r > MAX_RANK:
+                continue
+            gloss = clean_gloss(translation)
+            if not gloss:
+                continue
+            # On a duplicate headword keep the more frequent (lower-rank) entry.
+            if word in kept_rank and kept_rank[word] <= r:
+                continue
+            out[word] = gloss
+            kept_rank[word] = r
+
+    payload = json.dumps(out, ensure_ascii=False, separators=(",", ":")).encode("utf-8")
+    with gzip.open(dst, "wb", compresslevel=9) as gz:
+        gz.write(payload)
+    print(f"{len(out)} words -> {dst} ({len(payload)} bytes json)")
+
+
+if __name__ == "__main__":
+    if len(sys.argv) != 3:
+        sys.exit("usage: build_gloss.py <ecdict.csv> <out.json.gz>")
+    main(sys.argv[1], sys.argv[2])