Slide CRUD Phase 4 — Presentation.sections

Closes #11. Phase 4 of the slide CRUD epic. With this PR the seven sub-features in #11 are all shipped — Phase 1 (#33), Phase 2 (#34), Phase 3 (#35), and now Phase 4 here.

What this adds

Presentation.sections — a sequence-like collection over the p:extLst/p:ext{uri=521415D9-…}/p14:sectionLst extension that PowerPoint uses to organize slides into named groups in the slide pane.

prs = Presentation("deck.pptx")

# add a section
intro = prs.sections.add_section("Intro")

# move slides into it
intro.add_slide(prs.slides[0])
intro.add_slide(prs.slides[1])

# rename
intro.name = "Cover"

# read membership
[s.slide_id for s in prs.sections[0].slides]

# drop a section (slides remain in prs)
prs.sections.remove(intro)

Public API

• Presentation.sections → _Sections
• _Sections: __len__, __iter__, __getitem__, index(section), add_section(name, after=None), remove(section)
• Section: name (read/write), id (read-only GUID), slides (tuple), add_slide(slide), remove_slide(slide)

Membership invariants

Section membership references slides by their stable p:sldId/@id integer (NOT by r:id and NOT by position), so:

• Slides.move(...) — section assignment survives reorder ✓
• Slides.remove(...) — surviving slides keep their assignment ✓
• Slides.add_slide(layout, index=N) — new slides are unsectioned, existing assignments untouched ✓
• A slide can belong to at most one section. Section.add_slide(slide) automatically removes it from any other section first.

When a slide is removed from the presentation, its slide_id stays as an orphan in the section's p14:sldIdLst. Section.slides silently skips orphans on read; the XML round-trips them untouched. This deliberately matches python-pptx's "preserve foreign data" doctrine — we don't auto-prune.

PowerPoint compatibility

• Empty sections emit <p14:sldIdLst/> so PowerPoint versions that interpret an omitted sldIdLst as "all unsectioned slides" don't pull surprises.
• Section ids generated as {<UPPER-CASE-GUID>} matching PowerPoint's wire shape — e.g. {4080CFE4-F95C-449C-8898-95C81DD3D8B4}.
• Foreign p:extLst/p:ext siblings (p15:* modification tracking, etc.) round-trip untouched — the section ext lives alongside, not in place of.

Internal additions

• New pptx/sections.py module hosting Section and _Sections proxy classes.
• pptx/oxml/presentation.py extended with CT_PresentationExtensionList, CT_PresentationExtension, CT_SectionList, CT_Section, CT_SectionSlideIdList, CT_SectionSlideId, plus SECTION_LIST_EXT_URI constant and CT_Presentation.{section_list, get_or_add_section_list, remove_section_list} helpers.
• pptx/oxml/ns.py registers the p14 prefix (http://schemas.microsoft.com/office/powerpoint/2010/main).
• pptx/oxml/__init__.py registers the new element classes.
• p:extLst added as a successor entry on the existing ZeroOrOne sldMasterIdLst / sldIdLst / sldSz slots so insert ordering stays correct.

Tangential test-infra fix

tests/unitutil/cxml.py namespace-prefix grammar widened from Word(alphas) to Word(alphas, alphanums) so test fixtures can address p14:, w14:, o15:, etc. Local-name grammar already supported alphanums.

Test coverage

• 56 new unit tests in tests/test_sections.py:
  • oxml helpers (CT_SectionList, CT_SectionSlideIdList, the CT_Presentation section-list traversal helpers)
  • the Section / _Sections proxies
  • 8 round-trip integration tests including: no-sections baseline, names + membership, GUID preservation, membership-survives-move, removal pruning, orphan preservation through Slides.remove, empty-section round-trip, unicode + XML-special chars in section name, unsectioned-on-section-remove, sibling-ext preservation.
• 5 new behave scenarios in features/sld-sections.feature (default empty, add, slide membership, move-preserves-membership, remove cleans up extLst).
• UAT script uat_slide_sections.py (untracked per repo §6) builds a 6-slide / 3-section deck, demonstrates membership preservation through a slide move, and prints structural read-back. UAT signoff: ✓.

Out of scope (follow-up issues)

These were considered and deliberately left for separate PRs to keep this one reviewable:

• append_from cross-deck section porting: when source has sections and is append_from'd into a target, sections are not currently ported. The append_from machinery is Phase 3 territory; this is a worthy follow-up.
• PowerPoint-authored fixture round-trip: the round-trip tests use synthetic decks. A real PowerPoint-authored sectioned deck reopened and saved unchanged (XML-canonical-diff) would be the highest-confidence regression guard. Belongs in a manual UAT matrix, not unit tests.
• Per-slide / per-shape p:custDataLst tags: orthogonal to sections (different extension), tracked separately.

Verification

$ python3 -m pytest tests/ -q | tail -3
3222 passed in 4.33s

$ ruff check src tests | tail -3
All checks passed!

$ python3 -m behave features/ --no-color | tail -3
999 scenarios passed, 0 failed, 0 skipped
3000 steps passed, 0 failed, 0 skipped