docs: improve convention for handling static mut slots

[davidhewitt]

While considering CI failures in #5481 I realised our way of defining static "slots" is potentially questionable.

• We use a & immutable reference in the static, which seems questionable given the intent is to express mutable data. We cast this away using pointer casting.
• The reference seems to also requires us to read the mutable static's value to cast to a pointer, hence the unsafe block. Reading the value of mutable statics is known to be full of footguns.

The cleanest way I could find to do this, as proposed in this PR, is to store the mutable static as a static mut [T; N] array. It requires specifying N, because the compiler does not allow inference of the array size.

The nice thing about storing slots as an array is that we can then get a pointer to its first element with addr_of_mut!(SLOTS).cast(), which avoids any unsafe block or reading the value of the static.

[Icxolu]

I was having similar thoughts when I saw that failure yesterday. I think using an array here is much more appropriate and less sketchy. Having to type out the length is a bit annoying, but still quite readable I would say.

[davidhewitt]

Hmm annoyingly this looks like another thing blocked on #5486

[Icxolu]

That's unfortunate... I guess it's the reason we had the reference in there to begin with.