Emdash source with visual editor image upload fix

Fixes:
1. media.ts: wrap placeholder generation in try-catch
2. toolbar.ts: check r.ok, display error message in popover

packages/admin/src/lib/taxonomy-match.ts

@@ -0,0 +1,68 @@
+/**
+ * Taxonomy term matching for the admin picker.
+ *
+ * The picker filters an editor's typed input against existing terms. A naive
+ * `label.toLowerCase().includes(input.toLowerCase())` fails the accent case:
+ * typing `"Mexico"` does not substring-match a term labeled `"México"`
+ * because `"méxico".toLowerCase()` is still `"méxico"` and
+ * `"méxico".includes("mexico")` is `false`. The editor then sees zero
+ * suggestions and creates a duplicate `"Mexico"` term alongside the
+ * canonical `"México"`, splitting the taxonomy.
+ *
+ * This module folds diacritics via NFD decomposition before substring
+ * matching. No regexes are compiled from user input, so there is no ReDoS
+ * surface.
+ */
+
+const DIACRITIC_RANGE = /[\u0300-\u036f]/g;
+
+/**
+ * Case-fold + diacritic-fold normalization for substring matching.
+ *
+ * `"México"`, `"mexico"`, `"MÉXICO"` all collapse to `"mexico"`.
+ *
+ * NFD decomposes accented characters into a base + combining-diacritic
+ * sequence; the regex drops the combiners. Greek tonos, Vietnamese
+ * stacked diacritics, and other Latin-adjacent scripts are covered.
+ * Combining marks used meaningfully in non-Latin scripts (Arabic harakat
+ * U+064B–U+0652, Japanese dakuten U+3099) fall outside the U+0300–036F
+ * block and are left untouched — stripping them would change meaning.
+ */
+export function foldForMatch(value: string): string {
+	return value.normalize("NFD").replace(DIACRITIC_RANGE, "").toLowerCase();
+}
+
+/**
+ * Minimal shape a term must have to participate in matching.
+ * Kept structural so picker components and tests can use plain objects.
+ */
+export interface MatchableTerm {
+	label: string;
+}
+
+/**
+ * True if `input` is a substring of the term's label, ignoring case and
+ * diacritics.
+ *
+ * Empty or whitespace-only input returns `false` — the caller decides
+ * whether to show all terms or none in that state. The whitespace guard
+ * matters: without it, a needle of `"   "` would `.includes()`-match
+ * every term whose label contains a space.
+ */
+export function termMatches(term: MatchableTerm, input: string): boolean {
+	const needle = foldForMatch(input).trim();
+	if (!needle) return false;
+	return foldForMatch(term.label).includes(needle);
+}
+
+/**
+ * True if `input` is an exact (fold-equal) match for the term's label.
+ * Used to decide whether to show the "Create new term" button — if an
+ * editor types `"Mexico"` and a term labeled `"México"` already exists,
+ * Create must not appear or they'll produce a duplicate.
+ */
+export function termExactMatches(term: MatchableTerm, input: string): boolean {
+	const needle = foldForMatch(input).trim();
+	if (!needle) return false;
+	return foldForMatch(term.label).trim() === needle;
+}