chore(ariba): retire AribaPicker and _select_all_attachments (#195)

[alexwolson]

Closes #195.

Background

#185 removed AribaFileSource._read_expected_count — the last live caller of both AribaPicker
(the picker widget's Playwright adapter) and the module-level _select_all_attachments helper —
when the picker-count comparison it fed was proven to compare incommensurable quantities (54
picker vs. 39 traversal vs. 178 true leaves on one validated event). Neither was deleted at the
time, on the reasoning that removing working, hard-won code as a side effect of fixing a phantom
check wasn't that PR's call to make.

#186 then retired ariba_batch.py — the module AribaPicker was originally built to satisfy the
Picker protocol of — and deliberately left this pair alone too, per explicit instruction to track
their retirement as its own issue rather than fold it in.

What this does

Confirmed via grep before removing anything: no call sites for AribaPicker or
_select_all_attachments outside their own definitions and tests/test_ariba_select_all.py.

Removed as one cohesive cluster — everything orphaned by the same #185 cut:

• AribaPicker (the class itself, ~530 lines)
• _select_all_attachments, parse_total_mb, parse_selected_items, _selected_total_mb,
  _on_picker, _outline_sort_key — module-level helpers whose only callers were inside
  AribaPicker or the picker-count round trip
• _TOTAL_MB, _SELECTED_ITEMS, _ROW_KEY, MAX_BUNDLE_MB — regexes and a constant that only
  fed the functions above (the last one a leftover from the retired single-zip bundle ceiling
  check, referenced nowhere in source, only in the test file below)
• tests/test_ariba_select_all.py (148 lines, 11 tests)

Kept, because active code genuinely depends on them — traced every reference before touching
anything, not assumed:

• PICKER_HEADING — _on_event_view (still active, guards every traversal/download) checks for
  its absence to help confirm the content tree, not the picker, is in front of it
• _ListScroller and its four JS constants (_FIND_CONTAINER_BODY_JS, _CONTAINER_STATE_JS,
  _GEOMETRY_JS, _PROGRAMMATIC_SCROLL_JS) — genuinely shared scroll infrastructure, parametrised
  for both the picker and the content tree from day one. AribaFileSource is now its only user,
  but it was never picker-specific.

A gap this pass caught along the way

AribaFileSource's own class docstring still listed "Four things this class exists to get
right," with the fourth being "the count is read BEFORE the traversal, and the event view is
restored after" — the exact round trip #185 removed, never updated there. Fixed to three, and
updated the ~10 other historical comments across the file that referenced AribaPicker's specific
lessons (re-render invalidation, idempotent expansion, row ordering, the six-live-run scroll
debugging) to past tense with a #195 pointer, rather than leaving them read as though the class
they credit still exists.

Verification

• Every removal traced by grep before deleting — in particular, distinguished module-level
  helpers/constants used only by the orphaned cluster from ones shared with the still-active
  AribaFileSource/_ListScroller, since several sit interleaved in the same file.
• Module imports cleanly; confirmed via grep that no reference to AribaPicker,
  _select_all_attachments, or the other removed names remains anywhere in toronto_bids/ or
  tests/.
• 770 tests passing (781 − 11 deleted with test_ariba_select_all.py).

🤖 Generated with Claude Code

https://claude.ai/code/session_01W53WHx8mm2UHuLFAQWeF62