[GUIDE] Zero-Copy Out-of-Core Processing with Memory-Mapped Files (`mmap`)

[FrancescAlted]

A major capability of Blosc2 is processing datasets that far exceed physical RAM using mmap_mode="r" or "r+" in blosc2.open(). While mentioned in doc/guides/optimization_tips.md, there is no dedicated guide explaining how memory-mapped arrays interact with the operating system page cache, two-level chunking, and multi-threaded decompression.

Suggested Work

Create doc/guides/memory_mapped_storage.md and link it from doc/guides/index.rst.

The guide should cover:

1. The Out-of-Core Concept: How mmap_mode allows opening a 100+ GB array on a machine with 16 GB RAM with near-zero initial memory overhead.
2. Memory Profiling: Concrete code comparing standard reads vs. mmap_mode="r" with peak memory measurements.
3. Modes: Explaining read-only ("r"), copy-on-write ("c"), and read-write ("r+") semantics.
4. Chunk and Block Sizing for mmap: How block sizes affect decompression locality when paging from disk cache.
5. Concurrency with mmap: Concurrently reading memory-mapped containers across multiple processes.