Making explicit chunk cache handling policies and enabling a custom autochunker with a fall back to h5py's build-in autochunker

[mkuehbach]

Motivation:

• Customized chunking for compressed storage at the dataset level to allow tailoring the chunk layout better when slicing directions have a priori known usage biases towards a specific direction, like in EM, APM and other high volume techniques
• Making explicit the fact that internally chunking can be modified using buffer settings that take different effect on different hardware, like sequential or true parallel file systems. Currently configured with the default values that were hidden in the internals of the h5py library.

[RubelMozumder]

There are no tests for it. Multiple tests would be helpful to get the error before any unwanted breaks in production.

[mkuehbach]

There are no tests for it. Multiple tests would be helpful to get the error before any unwanted breaks in production.

I could add a test_chunk.py probing the return values of some instances of custom auto chunker

[mkuehbach]

There are no tests for it. Multiple tests would be helpful to get the error before any unwanted breaks in production.

I could add a test_chunk.py probing the return values of some instances of custom auto chunker

Tests added

[lukaspie]

Mostly good, just one additional comment. Thanks!

[lukaspie]

These three new files should be combined into one (e.g. chunk.py) since they all address the same functionality.

Alternatively, as discussed in #739, they should go to helpers.py.

[lukaspie]

This is a very large docstring, maybe we can have a link in the docs to this file to explain why/how the heuristic is different here.