fix: rewrite the test split path for yolov5 and yolov7 downloads

[roshaninfordham]

Description

Closes #156.

Version.__reformat_yaml prefixes the split paths in a downloaded data.yaml with the dataset location. For yolov5pytorch and yolov7pytorch it does this for train and val but not test:

if format in ["yolov5pytorch", "yolov7pytorch"]:
    content["train"] = location + content["train"].lstrip("..")
    content["val"] = location + content["val"].lstrip("..")

So the file that lands on disk carries two rewritten paths and one untouched one:

train: my-dataset-3/train/images
val:   my-dataset-3/valid/images
test:  ../test/images            # relative to the wrong directory

The mt-yolov6 branch directly above already rewrites all three.

Type of Change

• 🐛 Bug fix (non-breaking change which fixes an issue)

Changes Made

• roboflow/core/version.py — rewrite test alongside train and val for the yolov5/yolov7 branch.
• Guard it with if "test" in content. A version generated without a test split has no test key at all, and an unguarded assignment would turn a working download into a KeyError.
• tests/test_version.py — a TestReformatYaml case class covering yolov7 and yolov5, plus the no-test-split case.

Motivation and Context

I want to flag a decision rather than quietly make it.

#334 ("bugfix - downloading yolov8 files incorrectly amends data.yaml") took the position that this rewriting is itself the mistake — that the server's ../train/images is already correct and roboflow-python should leave it alone. That reasoning applies just as well to the yolov5/yolov7 branch, and removing the rewrite entirely is a defensible reading of this issue.

I did not do that, because it changes behaviour for every existing yolov5/yolov7 user and relitigates a decision that is yours rather than mine. What is not defensible under either reading is the current state, where one of three splits is treated differently from the other two within the same file. This PR makes the branch internally consistent and nothing more.

If you would rather delete the yolov5/yolov7 rewrite altogether, I am happy to send that instead — say the word and I will swap it.

Two other things I noticed and deliberately left alone:

• The mt-yolov6 branch indexes content["test"] unguarded, so it raises KeyError on a dataset with no test split. Same latent bug, different format; out of scope here.
• .lstrip("..") strips a set of characters, not a prefix, so it is equivalent to the .lstrip(".") used two lines above. Harmless for these inputs, but the two branches read as if they differ when they do not.

Testing

• I have tested this code locally
• I have added unit tests that prove my fix is effective
• All new and existing tests pass

Against unmodified main, two of the three new tests fail on exactly the reported symptom:

- ../test/images
+ /var/folders/.../tmpx2c0ibkb/test/images
FAILED (failures=2)

The third (no test split present) passes either way by design — it exists to pin the guard, so that a future change cannot reintroduce a KeyError.

python -m unittest     before: 970 tests, OK (skipped=1)
python -m unittest     after:  973 tests, OK (skipped=1)
ruff check roboflow/ tests/     All checks passed!
ruff format --check             2 files already formatted

Google Colab (optional)

Not applicable; the change is in local post-download file handling and is covered by unit tests that write and re-read a data.yaml in a temporary directory, with no API key or network access needed.