Add observability: pipeline run records, progress metrics, and actionable error surfacing

[szverev]

Problem

The pipeline is multi-step (Telegram fetch → extraction → embeddings → topics → scheduling → writing → publishing). When something fails or stalls, users rely on console logs, which makes it hard to:

• see what ran last
• understand what changed
• debug failures quickly

Goal

Introduce lightweight observability so users and contributors can diagnose runs and failures quickly.

Proposed approach

• Add a pipeline_runs table (or similar) that stores:
  • run id
  • start/end timestamps
  • step-level counts
  • failure summaries
• Emit structured logs (JSON optional)
• Add improved error surfacing:
  • include root exception messages (avoid swallowing)
  • include target URLs, status codes, and retry state

Acceptance criteria

• Each analyze/schedule/write/publish run is recorded
• CLI prints a concise summary + points to run id
• Failures include actionable error messages

Tasks

• Add DB schema + CRUD
• Instrument claws
• Add CLI command to show last N runs