LETKF on the monarch grid

[andreapiacentini]

Handling of the "member" key in the local I/O configuration (added by the LocalEnsemble solver).
Handling of date patterns in the I/O config.
A first test case

[andreapiacentini]

Tested with gnu compiler too.
Some issues can be foreseen for I/O efficacity with large ensembles and many variables in the States.
Other LETKF implementations have highly optimised this aspect. This investigation is related to Issue #22

[benjaminmenetrier]

I just cleaned a bit the code with commit 9978e08, @andreapiacentini let me know if it is fine for you.

[andreapiacentini]

It looks cleaner indeed.
What I still cannot understand is whether there is anything parallel in the I/O of the 4D LETKF.

[andreapiacentini]

Nor why the PseudoModel threatens the memory limitations by loading the whole timeseries of the states even if the forecast method advances step by step. Even with linear time interpolation you never need more that two instances in memory. And there's a loop on members too...