Add tests for bit manipulation intrinsics

[akiramenai]

LLVM can replace hand-written loops with bit manipulation intrinsics if loop idiom recognition pass decides so. There recent improvements in this pass that might affect performance and correctness after LLVM update. The goals of this ticket are to

• Ensure that we emit correct assembly for all bit manipulation intrinsics.
• Ensure that the code we generate for bit manipulations is no worse than a hand-written loop.