lib: graduate utf8.runesz and bufio.readbyte to (i32 | void)

Both used the -1 sentinel return; both had no external callers, so
the graduation is purely the API-shape change. utf8.runesz uses void
for "rune outside legal range"; bufio.readbyte uses void for EOF
(empty buffer). The full Hare shapes ((size | invalid) and
(u8 | EOF | io::error)) are still richer than this — those richer
returns arrive when utf8 grows an explicit invalid type and bufio
wires through io::stream's error path.

lib/bufio/bufio.ww

@@ -25,16 +25,17 @@ export fn init(b: *buf, s: streamp, data: *u8, cap: i32) void = {
 	b.w = 0;
 };
 
-// readbyte — pop one byte. -1 if empty. Hare name (transliterated
+// readbyte — pop one byte. void variant signals EOF (empty buffer).
-// from `read_byte`); the `-1`-for-EOF return is the sanctioned subset
+// Hare name; the (i32 | void) shape is a subset of Hare's full
-// of Hare's `(u8 | EOF | error)`.
+// (u8 | EOF | io::error) — error reporting from the underlying
-export fn readbyte(b: *buf) i32 = {
+// stream will arrive when bufio actually wires up to io::stream.
+export fn readbyte(b: *buf) (i32 | void) = {
 	if (b.r < b.w) {
 		let c: u8 = b.data[b.r];
 		b.r += 1;
 		return c: i32;
 	};
-	return -1;
+	return;
 };
 
 // Distinct alias so `(str | linerr)` has two variant types the

lib/encoding/utf8/utf8.ww

@@ -2,13 +2,14 @@
 // subset (no over-long encodings, no surrogates).
 
 def MAX:  rune = 1114111;	// 0x10FFFF
-def BAD:  rune = -1;
 
-export fn runesz(r: rune) i32 = {
+// runesz — encoded byte length of `r` as UTF-8. void variant means
-	if (r < 0)         { return -1; };
+// `r` is outside the legal range (negative, > 0x10FFFF).
+export fn runesz(r: rune) (i32 | void) = {
+	if (r < 0)         { return; };
 	if (r < 128)       { return 1; };
 	if (r < 2048)      { return 2; };
 	if (r < 65536)     { return 3; };
 	if (r <= MAX)      { return 4; };
-	return -1;
+	return;
 };