Implementing collect_meta(). Let's process per variable, so the first
step is to collect all meta and group per variable.

I'll collect everything here until I understand where it fits in the
library.

Understanding and validating the connections between zarr and MITgcm's
binaries. Once I validated a working reader, I can start to generalize
it.

Test driven development. Let's start with a hardcoded behavior and later
generalize it as I understand what is required.

Using lat as an easy case to start.

Still using a hardcoded case to understand the requirements for VarZ.

Trying a different concept and adding a few resources.

Chunk() is an auxiliary class to organize and collect information from
meta, while VarZ keep track of all chunks for one variable.

I was testing with @rabernat 's proof of concept.