fix: remove duplicate definitions shadowing imports and earlier methods

[harshadkhetpal]

Summary

Five dead duplicate definitions (ruff F811), each verified before removal:

1. core_steps/formatters/vlm_as_detector/v2_tensor.py — create_classes_index and scale_confidence are imported from gemini_detection_parsing at the top and then redefined locally mid-file. I diffed the local copies against the imported ones: byte-identical logic, so the local defs are pure dead weight (the shadowing means the import is what effectively runs anyway). Removed the local copies.

2. core_steps/models/foundation/pp_ocr/v1_tensor.py — InferenceConfiguration imported on two consecutive lines from inference_sdk and inference_sdk.http.entities (same class). Kept the public-path import.

3. core_steps/models/foundation/moondream2/v1.py — get_parameters_accepting_batches defined twice in the same class with identical return values; removed the shadowed first copy.

4. execution_engine/v1/dynamic_blocks/modal_executor.py and models/perception_encoder/vision_encoder/pe.py — redundant re-imports of datetime / nn shadowing module-level imports.

Deliberately left alone: the two describe_workflows_blocks FastAPI handlers in http_api.py (also F811) — routes register at decoration time so both endpoints work; renaming them is a cosmetic choice I'll leave to maintainers.

Testing

python -m py_compile passes on all five files; ruff check --select F811 on these paths goes clean. No behavior change — every removed definition was either shadowed before use or identical to what replaces it.

🤖 Generated with Claude Code

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[PawelPeczek-Roboflow]

would prefer to kill import in favour of contained blocks

[PawelPeczek-Roboflow]

LGTM apart from
inference/core/workflows/core_steps/formatters/vlm_as_detector/v2_tensor.py
I would prefer having local copies used to try to have blocks contained at the expense of code duplication.