x86/rounding-error test: add a bit of slack

[RalfJung]

@eduardosm either something is wrong with the rsqrt implementation, or with the test... when run with many seeds, it fails saying

correct result: 0.5, got: 0.49987793
that's an error of -0.00024414063 = 2^-11.999999

[rustbot]

Thank you for contributing to Miri!
Please remember to not force-push to the PR branch except when you need to rebase due to a conflict or when the reviewer asks you for it.

[eduardosm]

The relative error is exactly 2^-12, but calculating the log2 introduces some error, I believe that due to Miri introducing non-determinism to math functions. I think we should compare the relative error against 1.0/4096.0 and only use the log value in the assert message.

[RalfJung]

but calculating the log2 introduces some error,

🤦 of course. Thanks!