Incorporate memory model for strings and vectors to reduce overhead of size_type

[brawner]

These are my two main container classes my project uses, and I noticed they don't make use of the Memory Model concept. Can I add a request that this gets added for these classes so that the overhead of the size_type can be reduced?

[jwellbelove]

Did you see this as a global or per-instance memory model?
If it is per-instance, then there is a backward compatibility issue for etl::istring and etl::ivector.

[brawner]

I was thinking per-instance. Most uses of etl::string and etl::vector would be less than 256 in length for my project, but probably not all.

This request is primarily due to the overhead of etl::string<N> and etl::vector<char, N> for N < 256. Both would just need the char array and a uint8_t for its current size for its data members. So I was surprised to see that it would actually be 48 bytes. In the case of etl::string<N>, 8 bytes are required for the current_size and 8 for CAPACITY.

https://godbolt.org/z/nMsdzx5e4

After looking at the base class more closely, I see how backwards compatibility would be difficult to obtain with such a change. Perhaps another option would be to introduce small_string and small_vector classes.