ci: run miri + adjustments for miri

[phip1611]

Closes #117

[phip1611]

Could I ask you @nicholasbishop please for help? You do have more experience with miri than I have, I think. I don't quite understand why miri fails and if this is valid or how I should either fix or ignore it. Do you have some advice?

You can find a failed run here: https://github.com/rust-osdev/multiboot2/actions/runs/4526183512/jobs/7971182214

[nicholasbishop]

I think the error in multiboot2/src/boot_loader_name.rs:34 is due to rust-lang/unsafe-code-guidelines#256. I haven't read through that full thread, so there are probably more ways of handling this, but what I did in uefi-rs is convert the header type to a DST. So the BootLoaderNameTag.string field would become a [u8]. See https://github.com/rust-osdev/uefi-rs/blob/main/uefi/src/proto/device_path/mod.rs#L126 for one example of this.

[phip1611]

At this state, miri is somewhat useful. All tests that use the cast_tag() function (transitively) or the TagIter, need the #[cfg_attr(miri, ignore)] attribute.. I think that we are memory safe and miri just can't proof it.

A possible solution might be to model the whole parsing logic differently:

1. get pointer to mbi
2. find end
3. create a slice
4. operate on that slice. Then, all mem options are part of the "borrowed memory stack", according to miri

I need to think, if this should be merged in this state. I do not like the need for the (feature incomplete) AVec abstraction (aligned-vec).

[phip1611]

Hm. In #155 (comment) I could at least address some of the issues.

[phip1611]

This MR now introduces basic miri support. Most tests are skipped by miri. There are two major issues:

• Each test that uses the cast_tag (transitively) fails, as miri can't guarantee that the memory is valid, unfortunately
• when something like struct_as_bytes() ... cast is done, there are alignment issues as the returned Vec is of course only aligned to a one byte boundary... so far, in tests this never was a problem, luckily. At least on x86.

However, some things now are tracked by miri. For example if BoxedDst works properly.