refactor(db): dedupe cross-bundle wrappers into shared mixins (#1135)

[axisrow]

Part of #1135 (axis 5 of #1130). Stays off the parity surface — only genuine within-repo duplication is touched.

What

src/database/bundles.py duplicated identical repository pass-through methods across bundles:

• channel read/state ops — ChannelBundle + CollectionBundle
• task create / task read — both bundles
• get_setting/set_setting — four bundles (Collection, Notification, Search, Scheduler)
• account reads (list_accounts, list_account_summaries) — three bundles
• message search_messages/insert_messages_batch — Collection + Search
• get_recent_searches — Search + Scheduler

jscpd flagged 5 clone pairs inside the file — the top cluster of #1135.

How

Each duplicated wrapper group moves into a repo-scoped plain-class mixin (_ChannelOps, _CollectionTaskCreateOps, _CollectionTaskReadOps, _SettingsOps, _NotificationQueriesOps, _MessageOps, _AccountReadOps, _SearchLogReadOps). A bundle mixes in only the groups whose repository field it already carries.

Mixins are deliberately NOT dataclasses: their attribute annotations (channels: ChannelsRepository, …) do not become dataclass fields, so every bundle's positional constructor order is unchanged.

Verification

• Public API byte-identical — verified by an inspect.signature diff of all 10 bundle classes before/after. The only diffs are PipelineBundle.list_accounts/list_account_summaries, which gain the return-type annotations the duplicates already carried.
• Bundle method count 138 → 111; the test_db_access_conventions ratchet baseline is lowered accordingly (it is a downward-only cap by design).
• tests/test_bundles.py (42) + bundle consumers (scheduler, search_queries, notifier, photo_loader, pipeline_service, regression-services: 178) green.

jscpd

• Before (this file): 5 internal clones (~80 dup lines)
• After: 0 internal clones; the two remaining cross-file hits (bundles.py:19 ↔ facade.py:18 import block; bundles.py:734 ↔ photo_loader.py:456 repo pass-through) are import-boilerplate / single thin wrappers, not worth a helper.

Part of #1135 / #1130. Not for merge — review requested.

🤖 Generated with Claude Code

[axisrow]

🔍 Local review — cycle 1/3 (clean)

Reviewed locally via built-in /review + Codex companion (sol/xhigh), no GitHub bots pinged. Reviewed head 02decd1f against base 53001148.

This is a MOVE refactor (byte-identical relocation of pass-through wrappers into repo-scoped plain-class mixins). Both reviewers confirmed semantic equivalence:

Reviewer	Verdict	Findings
Codex (sol/xhigh)	approve	0 — "Dataclass constructor signatures unchanged, mixin annotations do not become fields, MROs contain no conflicting wrapper names."
/review (Claude)	clean	0 — equivalent MOVE; public API identical (verified by inspect.signature diff of all 10 bundle classes); ratchet baseline 138→111 lowered correctly.

No FIX, no UNVERIFIED, no SKIP — nothing to apply in cleanup. Round bound to 02decd1f; PR snapshot intact (author + head + base unchanged since round start).

Tests: test_bundles.py (42) + 6 consumer files (178) green after rebase. ruff clean.

Totals: 0 FIX, 0 SKIP, 0 UNVERIFIED. Review cycle complete — merge is yours to trigger (local mode does not auto-merge).

[axisrow]

🔍 Local review — cycle 1/3 (clean, post-rebase)

Rebased onto a main where #1287 (vulture #1136) already dropped three dead bundle methods, then reviewed locally via built-in /review + Codex companion (sol/medium), no GitHub bots pinged. Reviewed head 50ef5b80.

This is a MOVE refactor (identical pass-through wrappers relocated into 8 repo-scoped plain-class mixins). Both reviewers confirmed semantic equivalence after rebase:

Reviewer	Verdict	Findings
Codex (sol/medium)	approve	0 — "Effective bundle method sets and wrapper behavior match origin/main; dataclass field order is unchanged; set_active resolves to the correct repository for all five bundles; and the three #1287 removals remain absent."
/review (Claude)	clean	0 — verified by runtime inspect over MRO: method-set equivalent, dataclass fields unchanged (mixins are NOT dataclasses), set_active resolves correctly per bundle, #1287's 3 dead methods stay absent.

The subtle point both cleared: the set_active name is overloaded — ChannelBundle/CollectionBundle get _ChannelOps.set_active(pk) via the mixin, while AccountBundle(account_id), SearchQueryBundle(sq_id), PipelineBundle(pipeline_id) keep their own class-body definitions (no misresolution). Also: #1287 already removed list_live_usable_accounts/update_premium/get_message_stats, so this rebased PR carries 135→108 bundle methods (not the original 138→111); the ratchet baseline is lowered to 108 with a trace comment.

No FIX, no UNVERIFIED, no SKIP — nothing to apply in cleanup. Round bound to 50ef5b80; PR head + author intact.

Tests: test_bundles.py + ratchet + 6 consumer files (220) green after rebase. ruff clean. (Codex could not run pytest itself — its sandbox has no writable tmp dir — but the suite is covered locally.)

Totals: 0 FIX, 0 SKIP, 0 UNVERIFIED. Review cycle complete — merge is yours to trigger (local mode does not auto-merge).