Detect a live LiveTrack session from its last point, not session["end"]

session["end"].nil? was the active signal since this was first built,
on the assumption Garmin leaves it unset until the activity actually
stops. Checked directly against a real still-recording session today
(2026-09-20): end was already non-nil, set to exactly start + 24h — the
share link's own expiry, not an end time — so every currently-active
session was showing as "Ended". A genuinely ended session's end does
hold its real end time (confirmed against the 2026-09-06 session this
was originally, and only, verified against), which is what made the
bug easy to miss.

Recency of the last track point is a directly-observable signal
instead of reverse-engineering what a given `end` value means: Garmin
posts one roughly every postTrackPointFrequency seconds (~15, observed)
while genuinely still recording.

lib/maptool/livetrack_client.rb

@@ -1,5 +1,6 @@
 require "net/http"
 require "json"
+require "time"
 
 module Maptool
   # Fetches a Garmin LiveTrack session's track and metadata.
@@ -44,10 +45,22 @@ module Maptool
     # totals are cumulative for the whole session), name/type from the
     # session details — nil-safe throughout since either can be missing.
     #
-    # "active": Garmin doesn't set a session's `end` until the activity is
-    # actually stopped, so its absence means the session is still recording
-    # — inferred from the one (already-ended) session this was built
-    # against, not verified against a session actually caught mid-activity.
+    # "active": originally `session["end"].nil?`, on the assumption Garmin
+    # doesn't set `end` until the activity actually stops — disproven
+    # against a real still-recording session (2026-09-20): `end` was
+    # already non-nil, set to exactly `start + 24h`, the share link's own
+    # expiry, not an actual end time (a genuinely ended session's `end`
+    # instead reflects its real end — ~2h after start for the 2026-09-06
+    # session this was originally, and only, verified against). Recency of
+    # the last track point is a directly-observable signal instead of
+    # trying to reverse-engineer what a given `end` value means: Garmin
+    # posts one roughly every `postTrackPointFrequency` seconds (~15,
+    # observed) while genuinely still recording, so a point older than
+    # ACTIVE_WITHIN means the activity has actually stopped (or, at worst,
+    # lost signal for a while — same "looks ended until it isn't" behavior
+    # Garmin's own LiveTrack page shows in that case).
+    ACTIVE_WITHIN = 30 * 60 # seconds
+
     def self.summarize(result)
       last = result.points.last || {}
       session = result.session || {}
@@ -56,12 +69,20 @@ module Maptool
         "activity_type" => last["activityType"],
         "distance_meters" => last["totalDistanceMeters"],
         "duration_secs" => last["totalDurationSecs"],
-        "active" => session["end"].nil?,
+        "active" => active?(last["dateTime"]),
         "start" => session["start"],
         "end" => session["end"],
       }
     end
 
+    def self.active?(last_point_time)
+      return false unless last_point_time
+
+      Time.parse(last_point_time) > Time.now.utc - ACTIVE_WITHIN
+    rescue ArgumentError, TypeError
+      false
+    end
+
     private
 
     # The RSC payload serializes an absent field as the literal string