perf(lua): O(1) decode glyph lookup instead of brute-forcing 4 lengths

The decoder tried substring lengths 4,3,2,1 at every position — up to 4 string
allocations + 4 table lookups per byte. A glyph's byte length is fixed by its
UTF-8 leading byte (self-synchronizing; lengths never overlap), so utf8_char_length
gives it in one step. One substring + one lookup per position now.

Measured (random data, this machine, before -> after):
  decode 10MB:  7.77 -> 14.16 MB/s  (1.8x)
  decode  5MB:  5.49 -> 19.87 MB/s  (3.6x)
  decode  1MB:  9.62 -> 22.47 MB/s  (2.3x)
Lua test/test still passes (correctness preserved).

Author: pmarreck

bin/printable-binary

@@ -473,17 +473,17 @@ function PrintableBinary.decode(printable_string, options)
       matched_char = " "
     end
 
-    -- Try to match longest first (4-byte, then 3-byte, then 2-byte, then 1-byte)
+    -- A glyph's byte length is fixed by its UTF-8 leading byte (self-synchronizing),
+    -- so resolve it with ONE substring + lookup instead of brute-forcing all four
+    -- lengths (the map has no 4-byte glyphs, and lengths never overlap in UTF-8).
     if not matched_char then
-      for len = 4, 1, -1 do
-        if i + len - 1 <= s_len then
-          local sub = string.sub(cleaned_string, i, i + len - 1)
-          if decode_map[sub] then
-            result_bytes_as_chars[#result_bytes_as_chars + 1] = string.char(decode_map[sub])
-            i = i + len
-            matched_char = sub
-            break
-          end
+      local len = utf8_char_length(string.byte(cleaned_string, i))
+      if i + len - 1 <= s_len then
+        local sub = string.sub(cleaned_string, i, i + len - 1)
+        if decode_map[sub] then
+          result_bytes_as_chars[#result_bytes_as_chars + 1] = string.char(decode_map[sub])
+          i = i + len
+          matched_char = sub
         end
       end
     end