refactor: dedupe service/collector helper twins (#1135)

[axisrow]

Part of #1135 (axis 5 of #1130). Five genuine (non-parity) clones across services/ and telegram/.

Clones & extraction

Clone	Extracted helper	Notes
channel_analytics_service._calc_err ↔ _calc_err24	_engagement_rate(channel_id, rows)	Engagement formula sum / (rows*subs) * 100. Each caller keeps its own data fetch (get_err_data vs get_err24_data).
pipeline_nodes/handlers source block (LlmGenerate + AgentLoop)	module-level _build_source_messages(messages)	Identical [header] text (id:.. date:..) rendering, blank-joined; blank texts dropped. The AgentLoop user_message truthiness guard is preserved (joined string empty ⟺ parts list empty).
telegram_actions.download_media ↔ download_media_sized lookup	_fetch_message_with_flood_wait(...)	Resolve-entity + iter_messages(ids=) under flood-wait + MessageNotFoundError raise.
photo_task_service.send_now ↔ schedule_send batch+item ctor	_create_send_batch_item(..., status, schedule_at) → (batch_id, item_id)	Parameterized by status + started_at vs schedule_at.
collector_mixins/stats stats_*worker_count + available_stats_*worker_count (2 pairs)	_configured_worker_count, _available_worker_count	Twins differed only by config attr / default / log label.

Verification

• ruff clean on all five files.
• 1359 targeted tests pass (analytics/err/pipeline_node/telegram_action/download_media/photo_task/collector_stats/worker_count/source_messages).

jscpd

• Removed: the five clones above.
• Total: 0.67% (49 clones) — flat vs. the PR-2 state. The remaining services/telegram_actions.py:575↔594 and collector_mixins/cancellation↔stats hits are parity-style Telegram-error-handling blocks and the shared TYPE_CHECKING Collector Protocol (see tech-debt: дедуп кода (jscpd 0.56%, не-parity клоны) #1135 parity notes), intentionally untouched.

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/high — the first xhigh run exceeded the turn budget on this 5-file diff and was retried at high), no GitHub bots pinged. Reviewed head 36cfacab against base 53001148.

Five MOVE/extraction refactors — both reviewers confirmed semantic equivalence of each:

Reviewer	Verdict	Findings
Codex (sol/high)	approve	0 — "All five extractions preserve the relevant formulas, guards, exception behavior, timestamps, clamps, and fallbacks. The AgentLoop condition is equivalent because every appended source part is necessarily non-empty."
/review (Claude)	clean	0 — verified each: ERR formula + None-guards; _build_source_messages rendering identical (AgentLoop truthiness guard preserved — joined string empty ⟺ parts list empty); _fetch_message_with_flood_wait body 1:1; _create_send_batch_item field set matches (RUNNING→started_at, SCHEDULED→schedule_at); worker-count clamp + fallback identical.

The subtle point both flagged and cleared: the AgentLoop user_message guard changed from if source_parts (list) to if source_messages (joined string). Equivalent because the helper drops blank texts before joining, so every appended part is non-empty → the joined string is empty iff the parts list was empty.

No FIX, no UNVERIFIED, no SKIP — nothing to apply in cleanup. Round bound to 36cfacab; PR snapshot intact (author + head + base unchanged since round start).

Tests: 1373 targeted tests 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).