feat(utils): expose STANDARD_SECTIONS as the canonical section order

[katosh]

Expose the canonical section order as a public tuple alongside iter_outer so consumers that only need the names can read it directly — no generator, no per-section getattr, no backing-file toggling.

Closes #2401.

What changes

• New STANDARD_SECTIONS: tuple[AnnDataElem, ...] in src/anndata/utils.py — names the order iter_outer currently yields in.
• iter_outer now iterates STANDARD_SECTIONS instead of an inline list. Yield order, exception semantics, and the backing-file reopen/close behaviour are unchanged.
• Unit tests in tests/test_utils.py verify the constant matches the generator's yield order and that every name resolves to an attribute on a plain AnnData.

What does not change

• iter_outer's contract for every existing caller (AnnData._gen_repr, to_memory, _reduce). Same yields, same exception propagation.
• No new flags on iter_outer. Callers that want resilience against broken sections (the second motivation in Expose canonical section-name tuple alongside iter_outer #2401) iterate STANDARD_SECTIONS directly with their own per-section try/except.

Usage

from anndata.utils import STANDARD_SECTIONS

# Membership / ordering without driving the iterator or name-only listing
if section in STANDARD_SECTIONS: ...

Design notes

The section names are now declared in two places: the AnnDataElem Literal in src/anndata/_types.py (the type) and STANDARD_SECTIONS here (the order used by iter_outer). This is a deliberate trade, not a single source of truth. Alternatives considered:

1. Reorder AnnDataElem to match iter_outer's order; drop STANDARD_SECTIONS; iterate get_literal_members(AnnDataElem). This is genuinely one source of truth for both names and order. A grep of the repo shows no consumer of the current declaration order depends on it — ANNDATA_ELEMS in experimental/backed/_io.py is only used as **kwargs into AnnData(...) and as pytest.mark.parametrize values / filter(...) inputs. The cost is that AnnDataElem's member order becomes load-bearing: a stylistic reorder of the Literal silently changes iter_outer's yield order (and AnnData.__repr__ with it). Acceptable if documented with a comment and a test, but it overloads a typing construct with iteration semantics.

2. Replace AnnDataElem with a StrEnum. Also one source of truth, with native iteration. Changes the public shape of AnnDataElem (breaks typing.get_args consumers) — a larger blast radius than this PR warrants.

3. This PR: two declarations, consistency pinned by tests. The Literal stays a pure type (set of valid names), the tuple carries order. Names duplicated, but the concerns are decoupled — if AnnDataElem ever narrows (e.g., excludes X/raw/uns because they aren't aligned mappings), iter_outer's order doesn't silently break.

Happy to switch to option 1 if reviewers prefer the tighter coupling — it's a one-commit refactor. Flagging the trade rather than burying it in a single line.

Test plan

• pytest tests/test_utils.py — 4 passed (2 pre-existing, 2 new).
• Existing iter_outer call sites (AnnData._gen_repr, to_memory, _reduce) continue to work: the refactor is a pure loop-variable substitution.
• CI (pre-commit.ci + full test matrix).

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 85.48%. Comparing base (7861448) to head (dd15a5c).
✅ All tests successful. No failed tests found.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2402      +/-   ##
==========================================
- Coverage   87.43%   85.48%   -1.96%     
==========================================
  Files          49       49              
  Lines        7728     7729       +1     
==========================================
- Hits         6757     6607     -150     
- Misses        971     1122     +151     

Files with missing lines	Coverage Δ
src/anndata/utils.py	86.63% <100.00%> (-0.81%)	⬇️

... and 7 files with indirect coverage changes