Optimize IterKey

[rockeet]

IterKey is essentially an SSO(short string optimization), it can be optimized more, in this PR:

1. sizeof(IterKey) is reduce from 72 to 64 and inline buffer size is increased from 32 to 48
2. key_size_ and buf_size_ changed to uin32 because it is expected never larger than 2G
3. buf_size_ is also used as a flag to identify inline buffer or outline buffer
4. buf_ ptr is folded to be shared with space_ by using union
   • add getter func buf() makes the intention clear
5. is_user_key_ is changed to a bitfield which packed with buf_size_
   • buf_size_ is accessed less frequent than key_size_, use bit field will not harm performance

[ronag]

Make inline buffer smaller by 3 bytes and remove the nano optimizations (3,4,5) will make the code much more maintainable.

[rockeet]

Make inline buffer smaller by 3 bytes and remove the nano optimizations (3,4,5) will make the code much more maintainable.

SSO is really a bit tricky, but it is localized to IterKey internal, and IterKey is used widely in rocksdb, so efficiency should be prefered.

BTW: This code have been running for a long time in hour private branch, I just contribute our patch to the community. Also I have fixed the patch to pass the CI static analysis.