Add tensor storage format abstraction

[JanKleine]

Format abstraction is based on [https://doi.org/10.1145/3276493].

It adds a new Tensor data descriptor as subclass of Structure, which
automatically has the necessary fields for the relevant indices.

The following index types are currently supported:

• Dense
• Compressed (stores coordinates in segmented array akin to CSR)
• Singelton (stores one coordinate for every parent coordinate; used in COO)
• Range (computes coordinates based in current and parent dimension sizes + offset from parent)
• Offset (computes coordinates based on offset from parent coordinate)

[tbennun]

@JanKleine It looks nice. Since there is a lot of class infrastructure and code being added in this PR, I'd consider adding it to a new file (maybe dace.data as a folder).

Minor notes: there are some occurrences of sefl instead of self and Singelton should be spelled Singleton

[JanKleine]

Since there is a lot of class infrastructure and code being added in this PR, I'd consider adding it to a new file (maybe dace.data as a folder).

I tried to do that but circular dependencies seem to be a pain. Philipp also didn't know how to fix those easily. Do you (@tbennun) know if there is an easy way to do it? Otherwise I would postpone this to a different PR.

[phschaad]

LGTM, thank you for including the good doc comments and tests!