ci: fix fallback partition message, remove dead RETRY_CLEAN_CMD, fix precision flag

- run_parallel_benchmarks.sh: fix warning message to say gpu-rtx6000
  (not gpu-l40s) to match the actual fallback value
- phoenix/bench.sh: remove RETRY_CLEAN_CMD prefix on retry_build —
  retry-build.sh only uses RETRY_VALIDATE_CMD and always does rm -rf build
  on retry; variable was dead and re-introduced by upstream merge
- test.yml: fix --${{ matrix.precision }} expanding to bare '--' when
  precision is empty; restore PRECISION env var pattern; remove manual
  'Set up Python 3.14' step

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Author: sbryngelson

.github/scripts/run_parallel_benchmarks.sh

@@ -36,7 +36,7 @@ if [ "$device" = "gpu" ] && [ "$cluster" = "phoenix" ]; then
         fi
     done
     if [ -z "$BENCH_GPU_PARTITION" ]; then
-        echo "WARNING: No idle GPU partition found; falling back to gpu-l40s (may queue)"
+        echo "WARNING: No idle GPU partition found; falling back to gpu-rtx6000 (may queue)"
         BENCH_GPU_PARTITION="gpu-rtx6000"
     fi
     export BENCH_GPU_PARTITION

.github/workflows/phoenix/bench.sh

@@ -22,7 +22,7 @@ fi
 rm -rf build
 
 source .github/scripts/retry-build.sh
-RETRY_CLEAN_CMD="./mfc.sh clean" retry_build ./mfc.sh build -j $n_jobs $build_opts || exit 1
+retry_build ./mfc.sh build -j $n_jobs $build_opts || exit 1
 
 ./mfc.sh bench $bench_opts -j $n_jobs -o "$job_slug.yaml" -- -c phoenix-bench $device_opts -n $n_ranks