lex,wwi: \u/\U unicode escapes + wide-rune .wwi round-trip, both stages (#50)

Hare-faithful \u (4 hex) / \U (8 hex) escapes; \x/\u/\U share one codepoint
path (ref/hare/hare/lex/lex.ha lex_unicode); string literals UTF-8-encode
multi-byte codepoints (cstage inline utf8enc, wwstage utf8.encoderune). The
.wwi producer rune serializer now emits \u/\U so exported wide-rune defs
round-trip (was a fatal >0xFF). Reject >0x10FFFF and surrogates with Hare-
verbatim error strings. Closes the int-cast spelling divergence (#48 RUNE_MAX).
Both stages byte-identical; 446 tests pass.

cmd/w6c/wwi.c

@@ -167,14 +167,17 @@ wwi_quote(FILE *of, const char *s, u64 n)
 static void
 wwi_rune(FILE *of, u64 cp)
 {
-	/* The lexer's rune escapes stop at \xHH (no \u/\U — task #50), so a
-	 * codepoint above 0xff can't render as a re-parseable rune literal;
-	 * fail loud rather than emit a malformed one. No exported def names
-	 * such a rune today (RUNE_MAX is written as an int-cast for the same
-	 * reason). */
-	if (cp > 0xff)
-		fatal("wwi: rune codepoint U+%llx exceeds \\xHH (task #50)",
-		    (unsigned long long)cp);
+	/* #50: the lexer now reads \u/\U, so wide codepoints round-trip as
+	 * those escapes (write-twin of lex.c lexunicode). Codepoints <=0xff
+	 * keep the \xHH spelling they already round-tripped as. */
+	if (cp > 0xffff) {
+		fprintf(of, "'\\U%08x'", (unsigned)cp);
+		return;
+	}
+	if (cp > 0xff) {
+		fprintf(of, "'\\u%04x'", (unsigned)cp);
+		return;
+	}
 	unsigned char c = (unsigned char)cp;
 	fputc('\'', of);
 	switch (c) {