feat(coral-tpu): add compiled yolo26n EdgeTPU model + simplified Docker pipeline

Model files (compiled via ultralytics format='edgetpu'):
  models/yolo26n_full_integer_quant_edgetpu.tflite  (2.7 MB, Edge TPU)
  models/yolo26n_int8.tflite                        (3.7 MB, CPU fallback)

Docker / compile pipeline:
  - Simplify Dockerfile: python:3.11-slim + ultralytics + edgetpu_compiler
    Let ultralytics manage onnx2tf/TF versions (format='edgetpu' uses onnx2tf 1.28.8)
  - Simplify compile_model.py: single call to model.export(format='edgetpu')
    Per https://docs.ultralytics.com/guides/coral-edge-tpu-on-raspberry-pi/
  - Removes all manual ONNX→SavedModel→TFLite pipeline complexity

detect.py:
  - Updated _find_model_path to prefer *_full_integer_quant_edgetpu.tflite
    (ultralytics naming convention) over legacy *_edgetpu.tflite pattern

Author: solderzzc

skills/detection/yolo-detection-2026-coral-tpu/docker/Dockerfile

@@ -2,28 +2,21 @@
 # ─────────────────────────────────────────────────────────────────────────────
 # Coral Edge TPU Model Compiler
 #
-# Converts YOLO 2026 nano (.pt) → ONNX → TFLite INT8 → Edge TPU (.tflite)
+# Compiles YOLO 2026 nano to Google Coral Edge TPU .tflite format using
+# ultralytics' built-in format="edgetpu" export pipeline.
 #
-# IMPORTANT: Must run on linux/amd64. edgetpu_compiler only exists for
-# x86_64 Linux. On Apple Silicon or Windows, Docker Desktop handles emulation
-# via QEMU automatically — just pass --platform linux/amd64.
+# Per https://docs.ultralytics.com/guides/coral-edge-tpu-on-raspberry-pi/:
+#   model.export(format="edgetpu") handles the full pipeline:
+#   .pt → ONNX → onnx2tf SavedModel → TFLite INT8 → edgetpu_compiler
 #
-# Why tensorflow 2.13?
-#   edgetpu_compiler v16.0 only understands TFLite flatbuffer schema ≤ v3.
-#   tensorflow-cpu 2.13.1 is the last version that produces schema v3.
-#   TF 2.14+ produces schema v6 → "Invalid model" error from the compiler.
-#
-# Why NOT onnx2tf?
-#   onnx2tf ≥ 1.17 + TF 2.13 hits a TF-Keras symbolic-tensor TypeError
-#   during INT8 calibration. We bypass it entirely by using the ONNX Runtime
-#   directly to run calibration, then drive TFLiteConverter ourselves.
-#   This is the approach used in compile_model.py's two-stage pipeline.
+# MUST run on linux/amd64 — edgetpu_compiler is x86_64 Linux only.
+# On Apple Silicon or Windows, Docker Desktop handles QEMU emulation.
 # ─────────────────────────────────────────────────────────────────────────────
 
-FROM --platform=linux/amd64 python:3.9-slim
+FROM --platform=linux/amd64 python:3.11-slim
 
 LABEL maintainer="Aegis-AI / DeepCamera"
-LABEL description="Compiles YOLO 2026 nano to Google Coral Edge TPU .tflite"
+LABEL description="Compiles YOLO 2026 to Google Coral Edge TPU .tflite"
 
 # ── System deps ───────────────────────────────────────────────────────────────
 RUN apt-get update && apt-get install -y --no-install-recommends \
@@ -35,7 +28,7 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
         libglib2.0-0 \
     && rm -rf /var/lib/apt/lists/*
 
-# ── Google Coral edgetpu_compiler (x86_64 Debian package) ────────────────────
+# ── edgetpu_compiler from Google Coral apt (x86_64 only) ─────────────────────
 RUN curl -fsSL https://packages.cloud.google.com/apt/doc/apt-key.gpg \
         | gpg --dearmor -o /usr/share/keyrings/coral-edgetpu.gpg \
     && echo "deb [signed-by=/usr/share/keyrings/coral-edgetpu.gpg] \
@@ -45,31 +38,17 @@ RUN curl -fsSL https://packages.cloud.google.com/apt/doc/apt-key.gpg \
     && apt-get install -y --no-install-recommends edgetpu-compiler \
     && rm -rf /var/lib/apt/lists/*
 
-# ── Python deps ───────────────────────────────────────────────────────────────
-# Pipeline: ultralytics exports ONNX, then we convert ONNX→TFLite INT8
-# using onnx-tf (not onnx2tf) + tensorflow 2.13 + tflite-support calibrator.
-#
-# tensorflow-cpu==2.13.1  → TFLite schema v3, compatible with edgetpu_compiler v16
-# onnx-tf==1.10.0         → stable ONNX→TF SavedModel, works with TF 2.13
-# ultralytics             → YOLO .pt download + ONNX export only (not tflite export)
-# onnxruntime             → ONNX inference for INT8 representative dataset generation
-# numpy<2                 → required by tensorflow 2.13
+# ── Python: ultralytics handles all TF/onnx2tf version management ─────────────
+# ultralytics auto-installs: onnx2tf, tensorflow-cpu, onnxslim, etc.
 RUN pip install --no-cache-dir \
-        "tensorflow-cpu==2.13.1" \
-        "tensorflow_probability==0.21.0" \
-        "keras==2.13.1" \
-        "onnx-tf==1.10.0" \
-        "onnx>=1.14.0,<1.17.0" \
-        "onnxruntime==1.17.3" \
         "ultralytics>=8.3.0" \
-        "numpy>=1.24.0,<2.0" \
-        "Pillow>=10.0.0"
+        "numpy>=1.24.0,<2.0"
 
-# ── Copy compile script ───────────────────────────────────────────────────────
+# ── Copy compile entrypoint ───────────────────────────────────────────────────
 WORKDIR /compile
 COPY scripts/compile_model.py /compile/compile_model.py
 
-# ── Output volume ─────────────────────────────────────────────────────────────
+# ── Output volume (mount skill's models/ directory here) ──────────────────────
 VOLUME ["/compile/output"]
 
 # ── Entrypoint ────────────────────────────────────────────────────────────────