fix: truncate URI Template prefix modifier by character, not byte

Operator::inject() applied the ":N" prefix modifier with the
byte-oriented substr(), splitting multibyte UTF-8 characters in half
and producing invalid percent-encoded output (e.g. the euro sign
truncated to its first byte only).

RFC 6570 Section 2.4.1 defines the prefix length as a count of
characters, so use mb_substr() with an explicit UTF-8 encoding
instead. ext-mbstring is already a hard dependency of this package.

uri/UriTemplateTest.php

@@ -261,6 +261,20 @@ final class UriTemplateTest extends TestCase
         self::assertSame($expectedUri, $uriTemplate->expandToUrlOrFail($variables)->toAsciiString());
     }
 
+    public function testPrefixModifierTruncatesByCharacterNotByte(): void
+    {
+        // RFC 6570 Section 2.4.1: the prefix length counts characters, not octets, so a
+        // multibyte character must never be split. The euro sign and the musical G-clef
+        // are 3 and 4 bytes in UTF-8, respectively, but each is a single character.
+        $uriTemplate = new UriTemplate('{?currency:1}{clef:1}');
+        $variables = [
+            'currency' => "\u{20AC}uro",
+            'clef' => "\u{1D11E}stave",
+        ];
+
+        self::assertSame('?currency=%E2%82%AC%F0%9D%84%9E', $uriTemplate->expand($variables)->toString());
+    }
+
     public function testDisallowNestedArrayExpansion(): void
     {
         $template = 'http://example.com{?query,data*,foo*}';