CI: per-user mine timeout + lower max-api-fetches

At ~30-90 min cold-mine per user, an 80-user queue would take 40-100
hours. Bound each user to 25 min wallclock via 'timeout --kill-after
=30s 1500 python3 ...' so a single repo-heavy account can't block
the queue. Partial state is already deployed via the watcher's
interim deploys, so a timed-out user still shows up with whatever
phase it reached.

Also drop max-api-fetches from 1500 to 800. Most of the marginal
fetches are PR-detail / commit-stats that only matter for the rarer
"PR-only repo" lines counts — losing some completeness here is fine
for a first-pass prepopulate.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: pirate

.github/workflows/mine-and-deploy.yml

@@ -201,20 +201,27 @@ jobs:
             done
           }
 
+          # Per-user wallclock cap. Some users have hundreds of repos and
+          # cold-mining them takes forever; bounding to 25min/user keeps
+          # the queue moving (partial data already deployed via the
+          # watcher's interim deploys).
+          USER_TIMEOUT=1500
           while IFS= read -r user || [ -n "$user" ]; do
             user="${user%%#*}"
             user="${user//[[:space:]]/}"
             [ -z "$user" ] && continue
             [ "$user" = "pirate" ] && continue
             echo "::group::Mining @$user"
             # Run mining in the background; watch loop deploys partials.
-            python3 generate_stats.py --user "$user" \
+            timeout --kill-after=30s "$USER_TIMEOUT" python3 \
+              generate_stats.py --user "$user" \
               --no-search-commits \
-              --max-api-fetches 1500 &
+              --max-api-fetches 800 &
             MINE_PID=$!
             watch_and_deploy "$MINE_PID" "$user" &
             WATCH_PID=$!
-            wait "$MINE_PID" || echo "::warning::mining @$user exited non-zero"
+            wait "$MINE_PID" \
+              || echo "::warning::mining @$user exited non-zero (timeout or error)"
             # Stop the watcher and do a final deploy with the final HTML.
             kill "$WATCH_PID" 2>/dev/null || true
             wait "$WATCH_PID" 2>/dev/null || true