fix: have the campaign worker ask zl_bench for its plugin list

The worker carried a hand-copied plugin array, taken from zl_bench before
backends/ existed, and it had drifted twice over. It still named
_device_guard_plugin, which is now a shim for _device_setup_plugin. More
consequentially it carried no timing plugin at all, so every measurement this
campaign ever sealed was timer=direct while zl_bench's own driver had been
capturing -- a difference the harness never reported, because the missing plugin
cannot announce itself.

It now calls plugin_plan.py with ZL_BACKEND=musa and uses whatever comes back,
so the caliber follows the backend's declared capabilities instead of a snapshot
of them. The plan is preflighted for non-emptiness, since an empty array would
silently run pytest with no plugins at all: no shape injection, no autotune
record, no IR dump, and a table that still looks well-formed.

Two effects on this campaign. _timing_plugin now enters the list, and since
MUSA declares graph capture the worker will capture where the operator allows
it; operators that cannot are marked per row, as on any other backend. And
_parallel_warmup_plugin joins, inert unless PARALLEL_WARMUP_GPUS is set.

The sealed 32-workload run therefore sits on the far side of a caliber change
and is not comparable with anything produced after this commit. Re-running it is
a campaign decision, not a harness one.

The worker test asserted the literal "-p _musa_benchmark_compat_plugin"; it now
asserts the delegation and that no -p is spelled out locally. What the plan
holds for musa is covered where it belongs, in zl_bench's test_backends.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

campaigns/musa_rlc_product/run_pytest_musa_rlc_product.sh

@@ -53,17 +53,16 @@ export FLAGGEMS_PERF_CURRENT_OP="$OP"
 export FLAGGEMS_PERF_COLOR=never
 export NO_COLOR=1
 
-plugins=(
-  -p _musa_benchmark_compat_plugin
-  -p _device_guard_plugin
-  -p _seed_plugin
-  -p _shape_inject_plugin
-  -p _shape_iter_inject_plugin
-  -p _bespoke_shape_plugin
-  -p _autotune_record_plugin
-  -p _pretty_report_plugin
-  -p _ir_meta_plugin
-)
+# The plugin list comes from zl_bench, not from here. Hand-copying it is how
+# this campaign drifted: the copy was taken before backends/ existed, so it
+# still named _device_guard_plugin (now _device_setup_plugin) and, more
+# importantly, carried no timing plugin at all -- which is why every sealed
+# measurement is timer=direct while zl_bench's own driver had been capturing.
+# Asking the planner means the caliber tracks whatever the backend declares.
+export ZL_BACKEND=musa
+mapfile -t plugins < <("$PYTHON_BIN" "$ZL_BENCH_DIR/plugin_plan.py" \
+  --backend "$ZL_BACKEND" --pytest-args | tr ' ' '\n')
+preflight "plugin plan is non-empty" test "${#plugins[@]}" -gt 0
 
 dtype_args=()
 for dtype in ${DTYPES:-float16}; do

campaigns/musa_rlc_product/tests/test_musa_rlc_product.py

@@ -66,7 +66,13 @@ class MusaRlcProductTests(unittest.TestCase):
         self.assertIn(
             'FLAGGEMS_PERF_TTGIR_DUMP_DIR="$OUT_DIR/ttgir"', worker
         )
-        self.assertIn("-p _musa_benchmark_compat_plugin", worker)
+        # The plugin list is asked for, not written here: a hand-copied list is
+        # what let this worker keep _device_guard_plugin and no timing plugin
+        # long after zl_bench had moved on. What the plan contains for musa --
+        # the compat plugin included -- is asserted in zl_bench's test_backends.
+        self.assertIn("plugin_plan.py", worker)
+        self.assertIn('--backend "$ZL_BACKEND"', worker)
+        self.assertNotIn("-p _", worker)
         self.assertIn(
             'PYTHONPATH="$worker_dir:$ZL_BENCH_DIR:$FLAGGEMS_DIR/src:$FLAGTREE_SITE"',
             worker,