ci(lammps): retry tests after transient SIGTERM (#5804)

## Summary

- retry the LAMMPS pytest process up to three attempts only when it
exits with status 143 (`SIGTERM`)
- preserve all ordinary test failures and return their original exit
codes immediately
- handle the retry outside pytest because a process-level `SIGTERM`
prevents pytest plugins from recovering

## Background

The [failing
job](https://github.com/deepmodeling/deepmd-kit/actions/runs/29312634856/job/87019564217)
reaches the first active `PyLammps()` test and then exits without a
Python, LAMMPS, MPI, or pytest error:

```text
source/lmp/tests/test_lammps_pd.py
Process completed with exit code 143.
```

Exit 143 is `128 + SIGTERM`. The same failure signature has occurred
intermittently on unrelated commits and with different DeePMD backends.
Local fresh-process LAMMPS/MPI stress tests and repeated Paddle test
collection did not reproduce a deterministic DeePMD failure, so this
change treats it as a process-level transient while keeping actionable
failures visible.

## Validation

- `ruff check .`
- `ruff format . --check`
- `bash -n source/install/retry_on_sigterm.sh`
- verified an actual `SIGTERM` retries and succeeds on the next attempt
- verified a non-SIGTERM exit status is returned without retrying
- Paddle-only LAMMPS tests: `9 passed, 228 skipped`

Coding agent: Codex
Codex version: codex-cli 0.144.1
Model: gpt-5.6-sol
Reasoning effort: xhigh


<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->

## Summary by CodeRabbit

* **Bug Fixes**
* Improved automated test reliability by retrying interrupted test runs
caused by termination signals.
* Added validation and controlled retry limits to prevent unexpected
repeated executions.

<!-- end of auto-generated comment: release notes by coderabbit.ai -->

Co-authored-by: njzjz-bot <njzjz.bot@gmail.com>

Author: njzjz-bot

.github/workflows/test_cc.yml

@@ -76,7 +76,7 @@ jobs:
         if: matrix.enable_paddle
       - run: |
           export LD_LIBRARY_PATH=${{ github.workspace }}/dp_test/lib:$LD_LIBRARY_PATH
-          pytest --cov=deepmd source/lmp/tests
+          source/install/retry_on_sigterm.sh 3 pytest --cov=deepmd source/lmp/tests
         env:
           OMP_NUM_THREADS: 1
           TF_INTRA_OP_PARALLELISM_THREADS: 1

source/install/retry_on_sigterm.sh

@@ -0,0 +1,40 @@
+#!/bin/bash
+
+# Retry a command only when it was terminated by SIGTERM (128 + SIGTERM = 143).
+#
+# GitHub-hosted runners have intermittently terminated the LAMMPS pytest
+# process without a Python, LAMMPS, or MPI error.  Retrying only exit code 143
+# masks that external transient while preserving every actionable test failure.
+
+set -u
+
+if [ "$#" -lt 2 ]; then
+	echo "Usage: $0 MAX_ATTEMPTS COMMAND [ARG ...]" >&2
+	exit 2
+fi
+
+max_attempts=$1
+shift
+
+if ! [[ "$max_attempts" =~ ^[1-9][0-9]*$ ]]; then
+	echo "MAX_ATTEMPTS must be a positive integer." >&2
+	exit 2
+fi
+
+attempt=1
+while true; do
+	echo "Running command (attempt ${attempt}/${max_attempts}): $*"
+	"$@"
+	status=$?
+
+	if [ "$status" -eq 0 ]; then
+		exit 0
+	fi
+	if [ "$status" -ne 143 ] || [ "$attempt" -ge "$max_attempts" ]; then
+		exit "$status"
+	fi
+
+	echo "Command received SIGTERM (exit 143); retrying after 2 seconds." >&2
+	sleep 2
+	attempt=$((attempt + 1))
+done