perf(data-plane): avoid local batch copies

[rohitrango]

What does this PR do ?

Avoids unnecessary copies when fetching full or subset batches from the local data plane. It also preserves PackedTensor physical-segment deduplication across replica broadcasts so shared media is transferred once instead of being expanded per logical row.

Adds coverage that verifies full local fetches share tensor storage and replica broadcasts preserve PackedTensor row mappings, provenance, preprocessing configuration, and physical tensors.

This PR is stacked on NVIDIA-NeMo#4082 via `rohit/packedtensor_padding`.

Issues

None.

Usage

No user-facing API changes.

Before your PR is "Ready for review"

Pre checks:

• Make sure you read and followed Contributor guidelines
• Did you write any new necessary tests?
• Did you run the unit tests and functional tests locally? Visit our Testing Guide for how to run tests
• Did you add or update any necessary documentation? No documentation changes are needed.

Additional Information

Validated on Slurm job `18295320` (`pool0-01445`):

```text
uv run --frozen --group test pytest --confcutdir=tests/unit/data_plane -q tests/unit/data_plane/test_leader_broadcast.py
```

Result: 5 passed. Ruff passed for all three changed files, and Pyrefly reported zero errors for both implementation files.