Create public storage engine API for reducing impact of deadlocks during parallel replication

There are some corner cases where parallel replication can hit a
deadlock. This needs to be handled so that replication does not stall or even
fail.

In 10.0, a minimal patch was made, to solve the problem for InnoDB/XtraDB:
MDEV-5262, MDEV-5914, MDEV-5941, MDEV-6020.

In 10.1, we want to introduce a general storage engine API, properly
implemented as a service, that can be used by any storage engine to implement
the same optimisations/fixes.

Ideally, the API should be optional, and only used as an optimisation to
improve the performance or behaviour in case a deadlock does occur. So we
should try to find some way to handle deadlocks as well as possible for
storage engines that do not implement the API.

Mailing list threads that discuss the API and related issues:

https://lists.launchpad.net/maria-developers/msg07480.html
https://lists.launchpad.net/maria-developers/msg07489.html