Make the worker batch timeout configurable via LIDAR_BATCH_TIMEOUT

The pool of tile workers used to cancel every remaining tile after a
hardcoded 2-hour wall clock, silently truncating large batches (a
670-tile completion run lost its last 348 tiles that way). The timeout
now defaults to unlimited and can be capped per deployment with the
LIDAR_BATCH_TIMEOUT environment variable (seconds); the local worker
compose sets it to 6 hours.

💘 Generated with Crush

Assisted-by: Crush:glm-5.2

docker-compose.worker.yml

@@ -38,6 +38,8 @@ services:
       # Generations started from a remote map use the GPU
       - LIDAR_GPU=1
       - LIDAR_WORKERS=auto
+      # Wall-clock cap for one batch of tiles, in seconds (unset or 0 = unlimited)
+      - LIDAR_BATCH_TIMEOUT=21600
       # Workers per GPU capped by the free VRAM when the run starts
       # ((free - reserve) / per-worker peak); the excess runs on the CPU.
       # Peak estimated at 2048 MiB: adjust after measuring (nvidia-smi during a run).