Use tock-registers rather than volatile-register and clean up API

[qwandor]

This PR does quite a few things.

1. The biggest change is using tock-registers rather than volatile-register. tock-registers is used by the cortex-a crate, so it makes sense for users on Cortex A at least to have fewer dependencies. tock-registers itself has no dependencies, so we're trading three dependencies (bitflags, volatile-register and vcell) for one. It also makes for nicer and somewhat more typesafe code, as tock-registers has macros to generate types (including enums) for bit fields.
2. I've removed some of the public methods from the MmioUart8250 struct which were providing direct access to registers. If clients need functionality not provided by the wrapper methods then that functionality should be added to this crate, rather than having clients directly access registers.
3. I've marked MmioUart8250::new as unsafe. The caller needs to make sure that the address being passed in really is the base address of appropriate MMIO registers, and not an unmapped address, or the address of some other memory. Otherwise it would let safe code read and write arbitrary memory, which is unsound.
4. I added some tests.

[qwandor]

Oh right, and tock-registers does currently require nightly, but that should change when Rust 1.61 is released in May, which is planned to stabilise const_fn_trait_bound.

[duskmoon314]

Wow, this seems very reasonable and it is a breaking change.

I’m not familiar with tock-registers and I don’t know whether it is a better alternative to volatile-register. The rust embedded book and cortex-mare using volatile-register or its dependency vcell, so I used it at the beginning of this crate. Maybe it is better but I prefer a stable choice for this crate. Thus I would like to discuss this after the stabilization.

[qwandor]

Fair, I'll try to split out the other changes into a separate PR so we can get that in in the meantime.