All test cases use buffers that are multiples of the word size

[briansmith]

Consider this change to rdrand_exact():

@@ -91,7 +91,7 @@ unsafe fn rdrand_exact(dest: &mut [u8]) -> Result<(), Error> {
     let tail = chunks.into_remainder();
     let n = tail.len();
     if n > 0 {
-        tail.copy_from_slice(&rdrand()?[..n]);
+        unreachable!()
     }
     Ok(())
 }

I would expect at least one test to fail with this modification, but they all pass, because every test buffer is a multiple of the word size. We should have tests for this branch.

[josephlr]

Thanks! This is a good catch, we should add (a bunch) of tests with weird-length buffers.