spec: Maybe rip out alignment support?

[tv42]

snabbco/snabb#648 makes a case that unaligned access doesn't really cost much extra even for vector operations, these days (u variants of SIMD operations).

Perhaps that whole line of thinking is unnecessary these days, and we just eat the unaligned access cost.

That simplifies length-prefixed fields etc back to <len><content> pairs.

Should this prove the wrong decision, we can always add alignment support back later; no existing message would change.