Inline Index and IndexMut implementations

[valadaptive]

I ran into this when optimizing simdnoise. With gather ops no longer being a thing, I reimplemented them as standard loops. This led to a 20-30% slowdown in my application (which spends around half its time in simdnoise, so simdnoise itself is about 40-60% slower).

I thought the software gather ops were just inherently slower, but the actual problem is this line, with the innocuous-looking indices[i] access. i ranges from 0 to WIDTH, so the compiler should be able to easily remove the bounds check, but it couldn't, because the indexing operations couldn't be inlined.

With that fixed, simdnoise should now be as fast as it used to be before updating to simdeez 2.

[valadaptive]

@arduano Just wanted to make sure this (and my other PRs here) haven't fallen off your radar.

[arduano]

Apologies, will address all applicable PRs right now. I miss these notifications sometimes, and github does a terrible job at making sure I'm aware of them

[arduano]

I've published 2.0.0-dev5

Also just a note, the reason I'm still publishing dev versions is because I haven't ported across sleef. There's big architecture concerns around porting it, so I'm not sure if I should just bite the bullet and deprecate sleef support, or wait until someone PRs it in.