feat(weekly-report): Use recommended sort v1 to find top issues

[amy-chen23]

Resolves -----

Summary

• Replace the "top 3 issues by event count" approach with the recommended sort v1 algorithm (pure ClickHouse scoring) to find the top 5 issues for the weekly email report
• Use custom weights tuned for a weekly digest: heavier on event volume (0.40) and severity (0.30), lighter on recency (0.10), with spike zeroed out (not meaningful for a 7-day window — it measures a 6h/3d ratio at query time)
• Weights are registered as weekly-report.recommended.* options so they can be tuned via automator independently of the issue stream weights
• Add weight_overrides parameter to _recommended_aggregation() so the weekly report can pass its own weights without duplicating the ClickHouse expression logic
• Gated behind organizations:weekly-report-recommended-sort feature flag

Changes

• src/sentry/options/defaults.py — register 4 weekly report weight options
• src/sentry/search/snuba/executors.py — add weight_overrides param to _recommended_aggregation()
• src/sentry/tasks/summaries/utils.py — build candidates using _recommended_aggregation with weekly report weights
• src/sentry/tasks/summaries/organization_report_context_factory.py — populate candidates behind feature flag
• src/sentry/tasks/summaries/weekly_reports.py — simplify render path: use ClickHouse base_score directly, drop v2 reranking
• src/sentry/templates/sentry/emails/reports/body.html — conditional header: "Top issues for you" vs "Issues with the most errors"
• src/sentry/features/temporary.py — register feature flag
• src/sentry/snuba/referrer.py — add REPORTS_RECOMMENDED_ISSUES referrer
• tests/sentry/tasks/test_weekly_reports.py — 4 tests covering flag on/off, candidate ordering, template rendering

Test plan

• pytest -k recommended_sort tests/sentry/tasks/test_weekly_reports.py — 4 pass
• Enable flag for a test org, trigger weekly report, verify top 5 issues appear with "Top issues for you" header
• Verify flag-off path still shows "Issues with the most errors" with event-count ordering