sprint: mark data-dependent Day-4 tests with requires_data (Day 7 CI fix)

[Mark007-R]

What

Day 7 CI (added in #28) had two breakages from the same root cause: tests written before CI existed assumed the developer's local DVC state. PR #29 unblocked the install step (yanked protobuf). This PR unblocks the unit test step.

`pytest tests/` fails on CI because:

• `tests/test_api.py` boots the FastAPI app, which loads `models/fraud_model.pkl` and `data/processed/X_test.csv` -- both DVC-tracked, not in git.
• `tests/test_data_loader.py::test_loader_reads_x_test_and_y_test` reads the same two CSVs.

Fix

The CI workflow command already uses `-m "not requires_data"` to skip data-dependent tests -- the tests just weren't marked. This PR:

1. Adds `tests/conftest.py` registering the `requires_data` marker (so pytest doesn't warn about an unknown marker).
2. Tags `tests/test_api.py` at module level via `pytestmark = pytest.mark.requires_data` -- the whole file needs the model.
3. Tags only `test_loader_reads_x_test_and_y_test` in `tests/test_data_loader.py` -- the other two tests in the file (path resolution + dvc_status safe path) don't touch data and stay in CI.

Result

```
$ pytest tests/ -q -m "not requires_data" --ignore=tests/synthetic_drift.py
26 passed, 5 deselected in 76.74s
```

The 26 CI-runnable tests now cover: temporal-split regression guard, Pandas/Dask determinism, KS+PSI behaviour, registry promote/rollback end-to-end, retrain debounce, telemetry round-trips, loader path resolution, and DVC-status safe fallback.

The 5 deselected tests (`test_api.py` x4 + `test_loader_reads_x_test_and_y_test` x1) still run locally after `dvc repro train`.

Files touched

• `tests/conftest.py` (new, 18 lines) -- marker registration.
• `tests/test_api.py` -- 5-line comment + `pytestmark = pytest.mark.requires_data`.
• `tests/test_data_loader.py` -- import pytest + 1-line marker on the data-reading test.

No production code changes.