Add systemd units

Each repo carries the units for the daemons it owns. WorkingDirectory stays
/srv/astrape so env/ and models/ resolve, while the code comes from the
editable installs under repos/.

The old astrape-db.service was misnamed -- it runs the weather forecast
ingest, and collided with what would have been an astrape-db repo. It is
astrape-weather-forecast.service here.

The training unit forces ATEN_CPU_CAPABILITY=default, because this host has no
AVX2 and torch otherwise dies with SIGILL in Adam, and caps threads with Nice
so a nightly run does not starve the 5-second Modbus poll.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

deploy/systemd/astrape-usage-tcn-training.service

@@ -0,0 +1,22 @@
+[Unit]
+Description=Astrape Usage TCN Adaptive Training
+After=network-online.target
+Wants=network-online.target
+
+[Service]
+Type=oneshot
+User=gibil
+Group=gibil
+WorkingDirectory=/srv/astrape
+Environment=PYTHONUNBUFFERED=1
+Environment=PYTHONDONTWRITEBYTECODE=1
+# This host is a Xeon E5-2650 v2: no AVX2, no FMA. Torch dispatches kernels it
+# cannot execute and dies with SIGILL inside Adam, so force the scalar path.
+# The real fix is external_tcn on a GPU host; this keeps the local run alive
+# until the scheduler is pointed at it.
+Environment=ATEN_CPU_CAPABILITY=default
+# Training competes with a 5-second Modbus poll on 8 cores. Leave headroom.
+Environment=OMP_NUM_THREADS=4
+Environment=MKL_NUM_THREADS=4
+Nice=10
+ExecStart=/srv/astrape/.venv/bin/python -m astrape_oracle.scripts.tcn.schedule