Add aarch64 SIMD specialization

[llogiq]

I'm still without a good ARM CPU (but will hopefully get one, soon), but here's what I think should likely be the aarch64 intrinsics version of bytecount.

@Veedrac if you have a bit of time, I'd appreciate a review.

[llogiq]

There are still failing tests on aarch64. Notably the overflow_many tests fail for both count and num_chars. I'll look into that.

[Veedrac]

Am I good to wait for the tests to be fixed before reviewing, or would you prefer a review sooner?

[llogiq]

Yeah, I need to find a few hours. I think I know what's wrong, just need to fix things. I'll ping you.

[llogiq]

@Veedrac if all goes well, CI should be green soon. I've checked that perf matches the packed_simd variant on a M2 macbook and my mobile phone (using the bootstrap trick).

[llogiq]

@Veedrac CI is most certainly green. I'm tempted to just bump the version, push and publish it.

[llogiq]

I'm going to merge this now; it's fared well in all my tests, and if there's anything wrong, we can fix it in a followup PR.

[xtqqczze]

This apparently bumped MSRV to 1.61, see #101