Let callers keep an annotated image out of the Dataset

[iurisilvio]

What this does

POST /dataset/:id/annotate/:image adds the image to the project's Dataset unless the
request sends addToDataset=false, and save_annotation builds its URL without ever
sending that parameter:

url = f"{API_URL}/dataset/{project_url}/annotate/{image_id}?api_key={api_key}&name={name}"

So uploading an annotation for an image that is still waiting to be labeled or reviewed in
a batch pulls it out of that flow and into the Dataset, where the next dataset version picks
it up. A user hit this last week: a script annotated a batch through the SDK and 160 images
turned up in the project's Dataset with nothing approved. The Roboflow MCP server avoids it
by sending addToDataset=false explicitly; SDK callers had no equivalent.

add_to_dataset is now threaded through single_upload → Project.save_annotation →
rfapi.save_annotation → _save_annotation_url:

project.single_upload(
    annotation_path="labels.json",
    image_id=image_id,
    add_to_dataset=False,   # store the annotation, leave the image in its batch
)

Nothing changes for existing callers. The default is None, which omits the parameter
entirely and leaves the API's own default (add the image) in charge. Only an explicit
False/True puts addToDataset on the query string.

The endpoint's parameters are being documented on the API side in
roboflow/roboflow#16018.

Testing

TestSaveAnnotationUrl in tests/test_rfapi.py:

• omitting add_to_dataset leaves addToDataset off the URL, so existing callers upload
  exactly what they upload today;
• False and True each produce the matching query value;
• it composes with jobName, prediction and overwrite.

pytest tests/ --ignore=tests/manual — 1045 passed, 1 skipped. ruff format --check,
ruff check and mypy roboflow are clean (mypy reports the same 8 pre-existing
missing-stub errors as master).

🤖 Generated with Claude Code