Noah Stiltner

I have some seemingly working code for this, but it builds on the Rng PR with `variants.rs`. Would you like me to try to adjust it for the current state...

I took another look at my 64-bit counter code, and it isn't working correctly on the NEON backend atm. It uses a const bool in `variants.rs` to determine which way...

I've begun working on the second option... however... with the 1.60 MSRV, `Rudra` might not work on it since it is on nightly 1.58. I'll probably add the zeroize functionality...

I've made some progress with the second option, but even _before adding `zeroize`_, having a `union` for the `x86` backends seems to cause `avx2` performance to decrease by about 15%....

I benched a new branch (`zeroize_simd`) that essentially goes with the first option, just zeroizing after generation, and the `fill_bytes()` performance for avx2 ranged from 1.01 to 0.99 cpb, which...

Had a bunch of benchmarks here, but TL; DR: option 2 is more desirable now that it is working

Alright. Sorry for my complaining. I just didn't like that the first attempt at Option 2 resulted in Cipher's 1.6 cpb performance, even though it used pretty much the exact...

Would that be suitable for a `ChaChaCore` struct that contains a `union`? I've added the `ZeroizeOnDrop` code, but judging by the looks of your suggested implementation... it would be a...

Your ZeroizeOnDrop implementation seems to be far superior to a regular implementation of ZeroizeOnDrop. I've gone ahead and cleaned up my working branch a little and got it to pass...

Even though I made code for this issue, it seems that it would be a wasted effort to rework the backends given that typical constructor methods result in a stack-allocated...