AMOC scaleup attempts

[dcherian]

added watermark

1. MOC[{'transport_reg':1}] = MOC[{'transport_reg':1}] + moc_s this assignment is majorly slow. I replaced it with a concat statement
2. it's taking a while for the scheduler to go through the task graph
3. raise chunksize along nlon to 300, 1/3 the number of tasks now
4. DXU, DYU are float64, but UVEL etc are float32, this means iso_uflux, iso_vflux, wflux are all promoted to float64. Is that a good idea?

[TeaganKing]

@dcherian thanks for looking into this! Apologies I just saw this PR. That's helpful to reduce the number of tasks by increasing chunksize along longitude. I'm not entirely sure if we need float64 here, maybe @fredc has some insight on that.