feat: pre-cutover cleanup — smoke-test refresh, remediation→TelegramNotifier, coverage config (rc3)

[cipher813]

Summary

Three coordinated cleanups before the morning-signal consumer cutover. Bumps to 0.5.0rc3.

1. examples/smoke_test.py rewritten for the rc-era surface

Leads with FlowDoctor.builder() + TelegramNotifierConfig instead of the now-@deprecated flow_doctor.init() yaml flow. Adds smoke checks for flow_doctor.context() propagation, report_async() from an asyncio context, and flow_doctor.otel.report_to_otel_span_event. All offline (FLOW_DOCTOR_SKIP_PREFLIGHT=1 + fake creds + sqlite at temp path).

2. RemediationExecutor's orphan Telegram path migrated to TelegramNotifier

Previously RemediationExecutor had a bespoke _notify_telegram POST to an arbitrary webhook URL — a misnomer (Telegram doesn't have user-installable webhooks the way Slack does) and a duplicate code path that didn't compose with the rest of the notifier surface.

Added: TelegramNotifier.send_raw(text, *, parse_mode=, disable_notification=) for adjacent subsystems firing ad-hoc messages without conforming to the Report shape. Sentinel-based parse_mode / disable_notification lets callers distinguish "use instance default" from "explicit override".

Added to RemediationConfig: telegram_bot_token + telegram_chat_id + telegram_message_thread_id. _init_remediation in core/client.py builds a real TelegramNotifier from these (with the FLOW_DOCTOR_TELEGRAM_* env-var fallback chain) and hands it to RemediationExecutor via a new telegram_notifier= kwarg.

Deprecated: RemediationConfig.telegram_webhook_url — kept for 0.4.x yaml back-compat through the 0.5.x series, removed in 0.6.0. When both paths are configured, the notifier wins.

Remediation pings going through the new path pick up Markdown rendering, threading, target-id audit (actions.target row populated with the non-secret telegram:<chat_id>[:<thread>] identifier), and the same validate() preflight as the rest of the notifier surface.

3. Canonical coverage entry in pyproject.toml

pytest-cov was misreporting coverage on Pydantic-derived modules under editable installs because it instruments after the import has already happened — flow_doctor/notify/configs.py registered as 17% when it's actually 98%. The direct python -m coverage run -m pytest && python -m coverage report path measures correctly. Real project-wide coverage is 84%, not the 67% pytest-cov reported (still over the v0.5.0 plan's 80% target).

Also: .coverage artifact + htmlcov/ added to .gitignore.

Test plan

• pytest tests/ → 393/393 pass (376 prior + 17 new for the remediation-Telegram migration).
• python examples/smoke_test.py → all 10 smoke checks pass offline.
• Coverage measured via python -m coverage run -m pytest && python -m coverage report → 84% suite-wide, telegram.py 83%, executor.py 68% (was 56% — the migration tests lifted it).
• PyPI publish of 0.5.0rc3 after merge.
• Then morning-signal cutover in a separate PR (alpha-engine-morning-signal repo).

🤖 Generated with Claude Code