Add spike digest job with dedupe rules and report persistence

[artvandelay]

Motivation

• Provide a scheduled spike-reporting job that picks top spike events over a configurable window (e.g., 6h/24h) for downstream notifications.
• Prevent repeated or noisy alerts by applying dedupe rules based on last reported time, score delta, and new page spread.
• Produce a shareable digest in both Markdown and structured JSON containing term, score, direction, top pages, diff links, and snippets.
• Persist sent reports and supporting evidence so downstream systems can query history and avoid duplicates.

Description

• Added src/nlbt/spike_reports.py with dataclasses PageEvidence, SpikeEvent, and DedupeConfig and helper functions including select_top_spike_events, should_report_event, build_digest, ensure_reports_table, and run_scheduled_spike_digest to assemble and persist digests.
• Digest assembly outputs both Markdown and a structured JSON payload that includes term, score, direction, top_pages, diff_links, and snippets.
• Persisted reports into an SQLite reports table (created by ensure_reports_table) with stored payload_json and evidence_json for each reported item.
• Exported the new helpers from the package __init__ and added tests/test_spike_reports.py covering dedupe rules and report persistence.

Testing

• Ran pytest -q tests/test_spike_reports.py and it succeeded (2 passed).
• Running pytest -q for the whole repository failed during collection due to unrelated test collection issues (ModuleNotFoundError: No module named 'nlbt') and is unrelated to the new spike report unit tests.
• Unit tests exercise should_report_event, run_scheduled_spike_digest, and verify that data and snippets are persisted into the reports table.
• The new tests import the package from src during test execution so they run in the repository layout used here.

---

Codex Task