Ensure BPF read is 4-byte aligned

[frankplow]

Closes #641 and closes #645.

For pointer alignment, buffer_offset must be a multiple of 4. If it is not, the program will panic like in #641 and #645. This PR rounds buffer_offset up to the nearest multiple of 4 to avoid this issue. This also requires increasing the size of the buffer to ensure there is enough room to read into. I have made buffer_offset a member of DataLinkReceiverImpl as the allocation and use of the read buffer both require it, and previously it was derived twice in these two locations. By making it a member, it only needs to be derived once.

This is pretty hacky code. As the FIXME in the file indicates, there are much better ways of getting around this. I think this would require breaking API changes however, and I am not familiar with the codebase. For now I think this hotfix is sufficient.

[frankplow]

next_multiple_of was stabilised in 1.73.0. The MacOS runner is using an old stable version causing a build failure. It doesn't look like a strict MSRV is specified with a rust-version anywhere, so I'm not sure whether this is okay or not? I can move back to the old (x + align - 1) & !(align - 1) but that's pretty evil-looking code imo.

[tgross35]

Is it possible to run that arp_packet example in CI as part of the test suite?