[TheeCoderAhmed]

Fixes #1646.

For s3, gcs and azure stores, ObjectRef.full_path built <protocol>://<bucket>/<path> by hand and left out the store's location. The backend stores objects at <bucket>/<location>/<path> (StorageBackend._full_path), so on a store with a location the returned URI points where the object isn't: ref.fs.exists(ref.full_path) is False even though ref.read() works.

The cloud branches now return self._backend.get_url(self.path). That builds the URL from _full_path, the same path the backend writes to and fsmap already uses. The file branch and the fallback for other protocols are unchanged, so full_path still returns a plain filesystem path for local stores.

Added tests/unit/test_objectref.py. It covers s3 with and without a location, gcs and azure, using StorageBackend specs that don't touch the network. The three cases with a location fail on master (for example s3://bkt/_schema/... instead of s3://bkt/proj/outbox/_schema/...) and pass with the change. tests/unit/test_storage_urls.py and tests/unit/test_storage_adapter.py still pass, and ruff check / ruff format --check are clean on the changed files.

[dimitri-yatsenko]

1. Correct. StorageBackend.get_url goes through _full_path, which is where objects are written, so fs.exists(ref.full_path) now agrees with the backend.
2. A test with an s3 endpoint plus location, and a trailing-slash location ("proj/outbox/"), would pin the double-slash case if _full_path doesn't normalize it.
3. The file branch still builds its path by hand. It could also use get_url/_full_path for consistency, as a follow-up.