Zlib.gunzip should not fail with utf-8 strings

zstream_discard_input was encoding and character-aware when given input is user-provided, so this discards `len` chars instead of `len` bytes.

Also Zlib.gunzip explains in its rdoc that it is equivalent with the following code, but this doesn't fail for UTF-8 String.

```ruby
string = %w[1f8b0800c28000000003cb48cdc9c9070086a6103605000000].pack("H*").force_encoding('UTF-8')
sio = StringIO.new(string)
gz = Zlib::GzipReader.new(sio, encoding: Encoding::ASCII_8BIT)
p gz.read
gz&.close
```

ext/zlib/zlib.c

@@ -923,7 +923,7 @@ zstream_discard_input(struct zstream *z, long len)
 	    z->input = Qnil;
 	}
 	else {
-	    z->input = rb_str_substr(z->input, len,
+	    z->input = rb_str_subseq(z->input, len,
 				     RSTRING_LEN(z->input) - len);
 	}
     }

test/zlib/test_zlib.rb

@@ -1457,6 +1457,13 @@ def test_gunzip
       assert_raise(Zlib::GzipFile::Error){ Zlib.gunzip(src) }
     end
 
+    # Zlib.gunzip input is always considered a binary string, regardless of its String#encoding.
+    def test_gunzip_encoding
+      #                vvvvvvvv = mtime, but valid UTF-8 string of U+0080
+      src = %w[1f8b0800c28000000003cb48cdc9c9070086a6103605000000].pack("H*").force_encoding('UTF-8')
+      assert_equal 'hello', Zlib.gunzip(src.freeze)
+    end
+
     def test_gunzip_no_memory_leak
       assert_no_memory_leak(%[-rzlib], "#{<<~"{#"}", "#{<<~'};'}")
       d = Zlib.gzip("data")