Add per-page time-consumption statistics to flow analytics

[Copilot]

Flow analytics lacked visibility into which pages take the most student time. This adds per-page timing statistics derived from the gap between a page's last no-answer visit and its submitted-answer visit, with a toggle to include all sessions or only those counted in grade statistics (letting instructors distinguish TA time from student time).

Backend (course/analytics.py)

• New PageTimingStats dataclass: group_id, page_id, count, avg_time, min_time, max_time, stddev_time (minutes)
• New make_page_timing_stats_list(pctx, flow_id, include_all_sessions):
  • Correlated subquery finds the most-recent preceding no-answer visit per submitted-answer visit in the same session
  • ExpressionWrapper(DurationField) annotates elapsed time; negative durations filtered at DB level (clock-skew guard)
  • Count/Avg/Min/Max pushed to the DB via Django aggregation; sample stddev computed in Python (SQLite lacks STDDEV)
  • include_all_sessions=False restricts to participants with included_in_grade_statistics
• flow_analytics view accepts new timing_include_all_sessions GET param; passes page_timing_stats_list to template

Template (analytics-flow.html)

• New "Time Statistics Per Page" section: table sorted by mean time (most time-consuming first), columns: Page, Count, Min, Max, Mean, Std Dev (all in minutes)
• Toggle link switches between all-sessions and grade-stats-only; existing restrict-to-first-attempt toggle links preserve the new parameter

Tests

• Existing FlowAnalyticsTest mock-based tests updated to also patch make_page_timing_stats_list
• New MakePageTimingStatsListTest: empty result, no preceding view, basic stats, multi-visit math, sort order, session-filter correctness (asserts specific counts for student vs. TA participation)