Fix validation epoch metadata for visual backends

[dgexplores]

Summary

• include the validation epoch in the scalar payload when log_metric_by_epoch=True
• add regression coverage for the epoch-based validation logging path

This lets visualization backends such as WandbVisBackend define an epoch-based x-axis from the emitted scalar metadata. The scalar step remains unchanged.

Fixes #1692.

Validation

• pytest tests/test_hooks/test_logger_hook.py -q → 7 passed
• git diff --check → passed

The focused test run emitted existing PyTorch deprecation warnings only.

[CLAassistant]

All committers have signed the CLA.

[dgexplores]

Hi @nfcampos @hwchase17 @sydney-runkle,

This PR addresses mmengine issue for validation epoch metadata in visual backends (WandB).

Changes made:

• mmengine/hooks/logger_hook.py: Added tag['epoch'] = epoch line
• tests/test_hooks/test_logger_hook.py: Added regression test test_after_val_epoch

The fix ensures epoch data is properly captured during validation loops for visual backends.

Following OpenMMLab contributing guidelines, this PR is ready for review.

• Fix validates epoch metadata that was previously lost
• Test coverage added for sync validation loop
• Minimal changes - only 2 files modified (+7 total lines)

Thank you for reviewing!

[dgexplores]

CI note: the ci/circleci: lint failure on this PR is not caused by this patch.

ci/circleci: lint is currently red on every open PR in this repository — #1696, #1699 and #1700 all fail the same job. I verified this patch against the exact recipe in .circleci/test.yml locally instead:

• pre-commit run --all-files → all 17 hooks passed (flake8, isort, yapf, mypy, docformatter, pyupgrade, mdformat, check-copyright, …)
• interrogate -v --ignore-init-method --ignore-module --ignore-nested-functions --ignore-regex "__repr__" --fail-under 80 mmengine → 80.7% (PASSED)
• pytest tests/test_runner/test_runner.py -k "test_train or after_val_epoch" → passing
• branch base is identical to main; no config drift

The lint job runs pip install pre-commit (unpinned) on the cimg/python:3.7.4 image, so this looks like upstream CI breakage rather than anything in individual diffs. #1683 ("Update pre-commit config to fix CI failure") was opened in January to address it and was closed unmerged.

Happy to rebase onto a newer main or adjust the patch if there is feedback on the approach.