explicitly cast shifted constant to unsigned to avoid undefined behavior

[froydnj]

cc @Earlopain

I think this resolves the problem that #3681 was attempting to find -- shifting a signed integer in such a way that the sign bit changes is undefined behavior. So if token->type is 31, 1 << 31 is actually undefined.

Forcing the constant to be unsigned should fix this, and is good for consistency because the table we're masking against is full of unsigned integers anyway. I went ahead and changed the table initialization out of an abundance of caution, since we might have one of those token values be 31 someday; happy to undo that one if you like.

[Earlopain]

Thanks @froydnj! I don't have a super obvious way to test this, when the next version releases I can know for sure.

[froydnj]

Is it easy to figure out what compiler and version is being used in selenium-webdriver CI @Earlopain ? If you knew that (and ideally the compiler options they are using aren't too hard to figure out either), you could run smaller test cases through https://gcc.godbolt.org .

[Earlopain]

It's some bazel setup which I don't want to mess with too much. I gave a quick look through the build logs but nothing stood out to me. It can't be anything too exotic since I could swear I've seen this warning myself in the past.

[Earlopain]

Looks like this was exactly what was needed, their build goes fine now