sizeof(variant)

Nobody has claimed this yet.

My understanding is that calling `sizeof` on a variant should produce the size of the largest type the variant includes. And my assumption is that `sizeof` for a given type is not going to be consistent/portable across arches and platforms. However, as we should strive to have the variant use at little memory as possible I wanted to surface this ticket for discussion.

Dumb questions that I assume the answer is `duh, no`, but want to know for sure:
- Would it make sense to try to add tests of `sizeof(variant_instance)` by adapting to how `sizeof(std::string)` and other types might be different per platform? Might catch regressions if we ever made a mistake that increased the variant memory footprint.
- Are there any optimizations to be had from learning/applying ideas from http://www.catb.org/esr/structure-packing/?
- For custom types should we add some kind of MAX_VARIANT_SIZEOF flag to allow clamping the size - this would then be able to catch a situation were a poorly aligned custom type is larger than other built-in types used in the variant

Review the questions about sizeof(variant_instance), platform-dependent sizeof(std::string), structure packing, and custom types. Determine whether portable size-regression tests, packing changes, or a MAX_VARIANT_SIZEOF option are in scope; done requires a maintainer decision and corresponding tests or implementation.

Written by the indexing model from the issue text.