(refactor) Rename plan templates and add renderer experiment

- Rename benchmark_generator/plan.md to benchmark_plan.md
- Rename engineer/plan.md to engineering_plan.md
- Update todo.md references to match new plan filenames
- Update benchmark_plan_evidence_script.py templates for drafting mode
- Add renderer-parallel-modalities performance experiment

Co-authored-by: Qwen-Coder <qwen-coder@alibabacloud.com>

Author: MRiabov

scripts/experiments/performance/renderer-parallel-modalities/README.md

@@ -0,0 +1,61 @@
+# Renderer Parallel Modalities Experiment
+
+This experiment measures preview modality rendering on the current headless
+renderer backend. Parallel RGB/depth/segmentation fan-out is only supported in
+smoke-test mode; non-smoke runs stay sequential and fail closed if parallel mode
+is requested.
+
+It exercises the same `render_preview_scene_bundle(...)` code path that the
+renderer worker uses internally, but it runs on a synthetic scene so the result
+is not dominated by worker HTTP admission, retry loops, or seed bundle staging.
+
+The benchmark compares sequential rendering, and in smoke-test mode can also
+exercise the parallel branch.
+
+The default scene uses 24 views, all three modalities, axes and edges enabled,
+and a payload-path overlay, which matches the expensive part of the current
+preview workload.
+
+## Run
+
+From repo root:
+
+```bash
+uv run python scripts/experiments/performance/renderer-parallel-modalities/benchmark_parallel_modalities.py
+```
+
+Quick smoke test with a single view:
+
+```bash
+uv run python scripts/experiments/performance/renderer-parallel-modalities/benchmark_parallel_modalities.py \
+  --smoke-test-mode true \
+  --repetitions 1 \
+  --warmup-runs 0
+```
+
+Write to a custom output file:
+
+```bash
+uv run python scripts/experiments/performance/renderer-parallel-modalities/benchmark_parallel_modalities.py \
+  --output scripts/experiments/performance/renderer-parallel-modalities/results-custom.json
+```
+
+## Output
+
+The script writes a timestamped JSON result in this folder by default.
+
+Important fields:
+
+- `summary.parallel.mean_sec`
+- `summary.sequential.mean_sec`
+- `summary.parallel_minus_sequential_sec`
+- `summary.parallel_to_sequential_ratio`
+- `rows`
+
+How to read the result:
+
+- `parallel_minus_sequential_sec > 0` means parallel modality rendering was
+  slower on average.
+- `parallel_to_sequential_ratio < 1` means parallel modality rendering was
+  faster on average.
+- `rows[*].warmup=true` are intentionally excluded from the summary.