Change to a better way to define transactions

[shuaimu]

Currently a transaction in the system is written in dozens of hard-coded c++ functions, each of which is chopped by hand and contains a lot of function calls that should not appear in a transaction definition. This makes it very hard for maintenance and even harder to write new benchmark (even a much simpler one). This is probably why we did not manage to give the stock TPC-C benchmark before camera-ready. If we are going to do the stock TPC-C and other benchmarks such as TPC-E and SEATS in the future. So we need a proper way to write transaction definition before going any further!

I am thinking using coroutine plus macro-based (or other pre-compiling translation) annotations. This way we can make our codes neat, meanwhile we can still run the transaction efficiently in a C++ runtime. Moreover, it fits well with our single-execution model, giving us control at the possible blocking points (such as a lock in two-phase locking).

[shuaimu]

working on it.

1. First need to revise the txn2pl and piecestatus in memdb.
2. Next move all the TPL_KISS (and RCC_KISS) out of transaction definition.
3. Next refine all the variable definition in transaction into a map.
4. Wrap the control flow (chopper) hard codes into c++ macros.