docs(storage): settings-driven section prefixes, per-schema hash paths, GC semantics

[dimitri-yatsenko]

Documentation counterpart to datajoint/datajoint-python#1479 (in review), which aligned the code to the model these docs already described: the per-store hash_prefix / schema_prefix / filepath_prefix settings control the storage layout and are consumed by writers, garbage collection, and <filepath@> validation from the same spec values (no component-level fallbacks).

The load-bearing point (per maintainer review)

Hash-addressed paths embed the schema name — {hash_prefix}/{schema}/{hash} — which:

• scopes deduplication per schema (the type-system spec previously claimed "across the entire project" — corrected);
• lets GC attribute every stored object to a schema;
• and, because storage-side listing covers the whole section, means collect() must be given every schema that uses the store — the GC how-to's old note ("you only need to scan schemas that share the same database") asserted the opposite and is replaced with an explicit warning. Per-schema attribution is also what makes dropped-schema debris reclaimable.

Other corrections

• Layout parameterization: all path illustrations now show the configurable sections; schema-addressed pages no longer show the root-level {schema}/… layout (that's the ≤2.3.0 legacy layout, noted as such; GC lists both).
• Write tokens: schema-addressed filenames are {field}_{token}{ext} — several pages showed a dot-separated {field}.{token}.{ext} form that never matched the code.
• GC how-to: token-per-write versioning, folder objects + .manifest.json sidecars with coverage-based orphan matching, and the filepath_prefix never-touched guarantee.
• object-store-configuration spec: new normative Consumption paragraph (single source of values, defaults applied to every spec including plugin protocols), the hash-paths-embed-schema note, and the changing-prefixes-on-populated-store caveat (also added as a warning in configure-storage).

Sequencing

Merge after datajoint/datajoint-python#1479 — these pages describe the post-fix behavior. No file overlap with the open #195.

[dimitri-yatsenko]

Added a commit removing dj.gc.format_stats(stats) from the examples (garbage-collection.md, manage-large-data.md). That helper is being removed in datajoint/datajoint-python#1479 — it presumed a single display format and the library never called it. Examples now inspect the plain stats dict directly, and note that collect(dry_run=True) reports counts only (use scan() for per-item paths and reclaimable bytes).

[dimitri-yatsenko]

Added a commit updating the GC docs to the store-bound GarbageCollector class (datajoint/datajoint-python#1479): examples use dj.gc.GarbageCollector(store=...).scan(...)/.collect(...); the store binds at construction (no store_name= param). Two notable content changes: the Multiple Schemas guidance is now inverted — per-schema scoping makes any subset safe, so the old "always pass every schema or lose data" warning is replaced with a per-schema-attribution note; and the removed combined totals (orphaned/orphaned_bytes/referenced/stored) are dropped from the statistics reference (readers sum the per-section fields).