Implement backend caching and add GIF benchmarking and plotting tools

docs/performance.md

@@ -65,7 +65,10 @@ Against ~239 ms of inference, the difference is noise.
 ## Reproduce
 
 ```bash
-python scripts/bench_decode.py   # decode comparison
+pip install psutil matplotlib                          # bench/plot tools, not runtime deps
+python scripts/bench_gifs.py --procs "1" --duration 30  # single-worker profile -> bench_results.jsonl
+python scripts/bench_decode.py                          # decode comparison
+python scripts/plot_results.py bench_results.jsonl      # regenerate the charts in media/
 ```
 
 ## Results log

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "pyframe-gif-video-image-moderation"
-version = "0.2.0"
+version = "0.3.0"
 description = "Two-stage NSFW moderation for GIFs, videos, and images via local HuggingFace models and/or AWS Rekognition."
 readme = "README.md"
 requires-python = ">=3.10"

scripts/bench_decode.py

@@ -2,7 +2,7 @@
 """Micro-benchmark: cv2 file-path decode vs in-memory Pillow decode (the scan_bytes path).
 
 Shows the per-GIF decode cost of each, so you can see the millisecond delta against
-the ~789 ms ViT inference that dominates total time.
+the ViT inference that dominates total time (~91% of wall-clock; see docs/performance.md).
 
     python scripts/bench_decode.py [path/to.gif]   # synthesizes one if omitted
 """
@@ -52,4 +52,4 @@ def synth(path, n=60, w=320):
     print(f"  cv2 file-path decode:   median {c:6.1f} ms")
     print(f"  in-memory bytes decode: median {m:6.1f} ms")
     delta = m - c
-    print(f"  delta: {delta:+.1f} ms ({delta / c * 100:+.0f}%)  -- vs ~789 ms ViT inference, this is noise")
+    print(f"  delta: {delta:+.1f} ms ({delta / c * 100:+.0f}%)  -- vs the ViT inference (~91% of total), this is noise")