feat: paginate Strava following feed for 50–100 activities

[bin101]

Summary

• Bugfix/feature: Previously only ~13–16 activities were loaded because the JSON-XHR feed rewrite (commit 1bcfb83) dropped the quantity control of the old HTML parser and never read pagination.hasMore.
• Cursor pagination added: Strava's /dashboard/feed endpoint uses cursor-based pagination (verified via HAR capture, 2026-07-01). Each response entry carries cursorData: {updated_at, rank}; the next page is fetched with before=updated_at and cursor=int(rank) of the last entry.
• Load target: Loop runs while hasMore is true, up to _FEED_MAX_PAGES = 6 pages or _FEED_TARGET_ACTIVITIES = 100 unique activities.
• Human-like pacing: A random 0.5–1.5 s sleep between pages (RNG/sleep injectable for tests).
• No caller changes: engine.py and routes.py are untouched; FeedParser.parse() receives the same single-dict interface (merged pages stitched into one {"entries": [...], "pagination": {...}}).

Test plan

• TDD: 12 new tests written first, then implementation
• Multi-page merge (3 pages → all activities present)
• Stop on hasMore=false (exact 1 request)
• Stop at _FEED_MAX_PAGES cap (6 requests max even if hasMore stays true)
• First request has no cursor params (only feed_type + athlete_id)
• Second request carries before/cursor from previous page's last entry
• Deduplication (same activity id on two pages → appears once)
• Graceful stop when cursorData is missing despite hasMore=true
• Sleep called exactly n_pages – 1 times
• All 380 tests green, coverage 87% (≥85% target), ruff + mypy clean

🤖 Generated with Claude Code