[DOC] Container Architecture & Chunking Sizing Guide (Decision Matrix)

[FrancescAlted]

New contributors and users face two key architectural questions when getting started:

1. Which container should I choose? (NDArray vs CTable vs SChunk vs TreeStore vs DictStore vs BatchArray).
2. How do I choose chunk and block shapes? (Balancing compression ratio, CPU cache locality, slicing granularity, and storage overhead).

While issue #651 addresses a terminology glossary, there is no practical decision matrix or chunking rules-of-thumb guide in the documentation.

Suggested Work

Create a Markdown guide at doc/getting_started/choosing_containers_and_chunking.md (and link it from doc/getting_started/index.rst).

The guide should feature:

1. Container Decision Matrix: A structured table comparing:
   • Data Geometry (multidimensional, tabular/relational, key-value, hierarchical tree)
   • Read/write mutability and append performance
   • Query capabilities (expression evaluation, SQL/where filtering, coordinate slicing)
   • On-disk layout (.b2nd, .b2frame, .b2z, .b2d)
2. Chunking vs. Blocking Rules of Thumb:
   • Clear explanation of the two-level chunking hierarchy: chunks represent disk/network transfer units (typically 1 MB – 64 MB), whereas blocks represent CPU cache units (typically 64 KB – 1 MB to fit L2/L3 caches).
   • Practical guidelines for 1D, 2D, and 3D data geometries.
   • Trade-offs: large chunks (better compression) vs. small chunks (faster random slicing).