fix(playback): ignore single-letter words as Cyrillic encoding evidence

Codex review round 7 on #1471: a minimal CP1252 caption like "À la"
yields one lone Cyrillic-looking letter that passed both guards. Single
letters carry no trustworthy script information — an isolated accent
decodes to a lone Cyrillic letter, while genuine Cyrillic dialogue always
contains multi-letter words — so only words of length >= 2 now count as
Cyrillic evidence. Files whose only high bytes are isolated letters are
fundamentally ambiguous and default to Windows-1252.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

libs/ui/playback/src/lib/web-video-support/external-subtitle-cues.util.spec.ts

@@ -76,6 +76,21 @@ describe('external-subtitle-cues.util', () => {
             expect(decoded).toContain('À table');
         });
 
+        it('keeps a minimal isolated-accent CP1252 caption Latin ("À la")', () => {
+            // One lone Cyrillic-looking letter and two ASCII letters passed
+            // the earlier share guard; single-letter words must carry no
+            // script evidence at all.
+            const cp1252 = [
+                0xc0, // À
+                ...Array.from(
+                    ' la\n1\n00:00:01,000 --> 00:00:02,000\n'
+                ).map((c) => c.charCodeAt(0)),
+            ];
+            expect(decodeExternalSubtitleBytes(toBuffer(cp1252))).toContain(
+                'À la'
+            );
+        });
+
         it('keeps an isolated accented CP1252 word Latin ("À table")', () => {
             // À=0xC0 decodes under CP1251 to the pure-Cyrillic one-letter
             // word "А"; without the letter-share guard that single vote

libs/ui/playback/src/lib/web-video-support/external-subtitle-cues.util.ts

@@ -93,15 +93,20 @@ function chooseLegacySingleByteDecode(buffer: ArrayBuffer): string {
             asciiLetters += word.length;
         } else if (/[A-Za-z]/.test(word)) {
             mixedLetters += word.length;
-        } else {
+        } else if (word.length >= 2) {
+            // A single letter carries no trustworthy script information: an
+            // isolated CP1252 accent ("\u00c0"/"\u00c9") decodes to a lone Cyrillic
+            // letter and must not count as Cyrillic evidence, while genuine
+            // Cyrillic dialogue always contains multi-letter words.
             cyrillicLetters += word.length;
         }
     }
-    // Two guards: mixed-script words are strong evidence of misread Latin
-    // text, and isolated accented CP1252 words ("\u00c0 table" \u2192 "\u0410 table")
-    // masquerade as tiny pure-Cyrillic words \u2014 so Cyrillic must also carry a
-    // meaningful share of all letters before 1251 wins. Genuinely Cyrillic
+    // Two guards on top of that: mixed-script words are strong evidence of
+    // misread Latin text ("\u00e9tait" \u2192 "\u0439tait"), and Cyrillic must carry a
+    // meaningful share of all letters before 1251 wins \u2014 genuinely Cyrillic
     // dialogue dominates its own letter count even with embedded Latin names.
+    // A file whose only high bytes are one or two isolated letters is
+    // fundamentally ambiguous; defaulting it to 1252 is the best effort.
     const plausiblyCyrillic =
         cyrillicLetters > mixedLetters * 2 &&
         cyrillicLetters * 4 > asciiLetters;