Add v2 trainings + train-recipe support (describe, non-blocking start_training, CLI)

[leeclemnet]

What does this PR do?

Adds SDK + CLI support for training with a custom trainRecipe (hyperparameter/augmentation sweeps) against the v2 trainings API.

• Adapters (roboflow/adapters/rfapi.py): get_train_recipe (GET .../v2/trainings/recipe?modelType=), create_training (POST .../v2/trainings, returns {"trainingId", ...}), list_version_trainings, get_version_training — same URL/auth/error conventions as the existing train adapters.
• SDK (roboflow/core/version.py): Version.describe_train_recipe(model_type) (schema + ready-to-submit template), non-blocking Version.start_training(...) (builds a full recipe from the server template when hyperparameters/online_augmentation are given; mirrors train()'s generation-wait + export-ensuring; docstring includes a sweep-loop example), Version.list_trainings(), Version.get_training(). The blocking Version.train() is untouched.
• Merge helper (roboflow/util/train_recipe.py): build_train_recipe — shared by SDK and CLI so both submit identical recipes; folds a top-level epochs into the recipe's hyperparameters unless explicitly set there (the server prefers the dense-filled recipe epochs over body epochs, so without the fold a caller's epochs= would be silently ignored).
• CLI: roboflow train recipe -p <project> -v <N> -m <model_type> (prints schema+template JSON) and --hyperparameters '<json>' / --train-recipe '<json>' on roboflow train start (v2 path, prints the new trainingId; clean errors on invalid JSON / conflicting flags).

Related Issue(s): Companion to roboflow/roboflow#13699 (the backend describe endpoint — this PR's describe_train_recipe/--hyperparameters paths require it; deploy order: platform first, then release this). MCP companion: roboflow/roboflow-mcp#120.

Type of Change

• New feature (non-breaking change that adds functionality)

Testing

• I have tested this change locally
• I have added/updated tests for this change

Test details:
+46 tests: responses-based adapter tests (URLs, query params, conditional camelCase body keys, error raise), Version tests with patched adapters (template merge, epochs fold incl. no-clobber and aug-only cases, export-ensuring), CLI tests via CliRunner + patched adapters (recipe command happy/error paths, --hyperparameters merge, --epochs fold, conflicting-flags error). Full suite: python -m unittest → 909 tests OK. ruff format --check, ruff check, mypy roboflow all clean. CLI smoke-tested via the installed entry point.

Checklist

• My code follows the style guidelines of this project
• I have performed a self-review of my own code
• I have commented my code where necessary, particularly in hard-to-understand areas
• My changes generate no new warnings or errors
• I have updated the documentation accordingly (if applicable) — CLI-COMMANDS.md updated; docs.roboflow.com CLI reference is a tracked follow-up

Additional Context

Sweep usage the PR enables:

v = rf.workspace().project("safety-vests").version(5)
ids = [v.start_training(model_type="rfdetr-nano", hyperparameters={"lr": lr})["trainingId"]
       for lr in (1e-5, 1e-4, 1e-3)]

Recipe-based training requires the workspace's online-augmentation training entitlement (403 otherwise). Against a backend without the describe endpoint, describe_train_recipe/knob kwargs fail with the upstream 404; passing a full train_recipe works today.

🤖 Generated with Claude Code