recipe(yolos-fashionpedia): add object detection recipes

[kujin66]

Adds recipe coverage for valentinafevu/yolos-fashionpedia as a YOLOS object-detection model. The current origin/main baseline already exports the checkpoint, so this is an L0 recipe-only contribution with measurable L1 value from the checked-in w8a8 quantized recipe running successfully on CPU.

1. Recipe path(s)

• examples/recipes/valentinafevu_yolos-fashionpedia/object-detection_fp16_config.json
• examples/recipes/valentinafevu_yolos-fashionpedia/object-detection_w8a8_config.json

2. README row

• Added valentinafevu/yolos-fashionpedia | object-detection to examples/recipes/README.md.
• Updated total from 75 to 76.

3. Build output dir

• Baseline: temp/baseline_yolos_fashionpedia
• Recipe fp16: temp/yolos_fashionpedia_fp16_recipe2
• Recipe w8a8: temp/yolos_fashionpedia_w8a8_recipe
• L1 perf JSON: temp/yolos_fashionpedia_w8a8_perf_cpu.json

4. Build log

• Baseline command: uv run winml build -m valentinafevu/yolos-fashionpedia -o temp/baseline_yolos_fashionpedia --ep cpu --device cpu --no-analyze --no-optimize --no-quant --no-compile --rebuild
• Baseline result: Build complete in 42.6s; final artifact temp/baseline_yolos_fashionpedia/model.onnx.
• Recipe fp16 command: uv run winml build -c examples/recipes/valentinafevu_yolos-fashionpedia/object-detection_fp16_config.json -m valentinafevu/yolos-fashionpedia -o temp/yolos_fashionpedia_fp16_recipe2 --no-analyze --no-compile --rebuild
• Recipe fp16 result: Build complete in 34.4s; FP16 stage ran in 1.9s, artifact 58.6 MB, 226 FLOAT16 initializers, I/O types preserved as FP32.
• Recipe w8a8 command: uv run winml build -c examples/recipes/valentinafevu_yolos-fashionpedia/object-detection_w8a8_config.json -m valentinafevu/yolos-fashionpedia -o temp/yolos_fashionpedia_w8a8_recipe --no-analyze --no-compile --rebuild
• Recipe w8a8 result: Build complete in 71.1s; quantized artifact 30.0 MB.

5. Appended findings

• No in-repo model_knowledge file is included in this Lane B model PR. Model-specific finding to record externally: yolos: valentinafevu/yolos-fashionpedia resolves as YolosForObjectDetection/object-detection; input image shape is [1,3,512,864]; fp16 and w8a8 CPU recipes build, and w8a8 runs perf.

6. Optimum-coverage probe

{
  model_type: yolos,
  architectures: [YolosForObjectDetection],
  vendor: [],
  after_winml: [feature-extraction, object-detection],
  added_by_winml: [feature-extraction, object-detection]
}

7. Claimed (Effort, Goal, Outcome)

• Effort: L0 recipe-only, using existing WinML YOLOS registration.
• Goal ceiling: L1, because baseline already covers L0 and the recipe delta is quantized w8a8 CPU perf evidence.
• Outcome: L0, recipe JSON plus README row only.
• Target EPs: cpu.
• Baseline freshness: origin/main and baseline HEAD were both 904c20109ab31d737c5132c3686ab404236edd57.
• PR head: e5cd3ec3.
• winml --version: 0.2.0.

8. Goal-ladder verdict table
   | Tier | Verdict | Evidence |
   |---|---|---|
   | L0 | PASS | WinMLBuildConfig.from_dict(...).validate() passed for both recipes; fp16 recipe build completed in 34.4s and emitted 226 FLOAT16 initializers; w8a8 recipe build completed in 71.1s; ONNX input shape is pixel_values [1,3,512,864]; outputs are logits [1,100,47] and pred_boxes [1,100,4]. |
   | L1 | PASS | CPU EP perf on temp/yolos_fashionpedia_w8a8_recipe/model.onnx completed with CPUExecutionProvider; providers snapshot was [DmlExecutionProvider, CPUExecutionProvider]; mean latency 558.923 ms, p50 560.175 ms, throughput 1.79 samples/sec, RSS total delta 305.65 MB; perf JSON saved to temp/yolos_fashionpedia_w8a8_perf_cpu.json. |

9. Methodology-evolution declaration

• No new methodology friction observed for this model PR. The earlier mixed-branch state was corrected by applying the existing _meta-044/Lane B branch-isolation rules, so no new _meta-NNN or skill-file edit is required.
• No skill files are edited in this Lane B model PR.

[kujin66]

Reviewer-style verification (not a formal GitHub approval; same account authored the PR):

• PR branch head: e5cd3ec33bfaa14eaf037be937b720c394810274.
• Scope check: diff is limited to examples/recipes/README.md plus the two examples/recipes/valentinafevu_yolos-fashionpedia/*.json recipe files. No source, tests, or skill files are included.
• Baseline freshness: PR body cites origin/main/baseline HEAD 904c20109ab31d737c5132c3686ab404236edd57; current origin/main was refreshed and matches that commit.
• Recipe check: object-detection_fp16_config.json uses quant.mode=fp16; rebuilt artifact temp/yolos_fashionpedia_fp16_recipe2/model.onnx has 226 FLOAT16 initializers and preserved I/O. object-detection_w8a8_config.json artifact contains QDQ nodes (QuantizeLinear=440, DequantizeLinear=666).
• L0/L1 evidence: PR body includes build output dirs, build-complete lines, structural I/O shapes, CPU provider snapshot, and w8a8 perf numbers.
• CI: all checks are passing at time of comment: license/cla, lint, Analyze (Python), CodeQL, and all test jobs.

Remaining gate: external maintainer/reviewer approval and conversion from Draft to Ready.