[BugFix] Use torch.nn.Buffer instead of the tensordict placeholder

[ahlag]

Description

On torch>=2.5, tensordict.utils.Buffer is an empty placeholder class, so isinstance(x, Buffer) is always False and the buffer branches in _cast, _map_weight and _to_module_compatible_tensor never run: a torch.nn.Buffer comes back as a plain tensor.

• torchrl/_utils.py now imports torch.nn.Buffer, falling back to the tensordict copy on torch<2.5 (same try/except ImportError pattern as is_compiling just above it). collectors/utils.py, weight_update/weight_sync_schemes.py and objectives/common.py import Buffer from there.
• _make_target_param checked isinstance(x, nn.Parameter) after taking x.data, which is never a Parameter. It now checks the input. This doesn't change what users see, since TensorDictParams already registers the target params as buffers, but the branch now does what the comment above it says.
• test_weight_helpers_keep_buffers in test/test_collectors.py covers the three helpers (fails on main, passes with the fix; skipped on torch<2.5, where nn.Buffer doesn't exist).

Tested on macOS (CPU), torch 2.14.1, tensordict 0.14.3: the new test plus 129 related collector/loss tests, and test/objectives/test_loss_module.py, test_dqn.py, test_ddpg.py (1491 passed). ufmt and flake8 clean at the pinned pre-commit versions. I couldn't run torch<2.5 locally, so could a maintainer add the ci/olddeps label?

I used an AI assistant (Claude Code) to help write this change; I've reviewed and tested it.

Motivation and Context

close #4535

• I have raised an issue to propose this change (required for new features and bug fixes)

Types of changes

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

Checklist

• I have read the CONTRIBUTION guide (required)
• My change requires a change to the documentation.
• I have updated the tests accordingly (required for a bug fix or a new feature).
• I have updated the documentation accordingly.

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/rl/4541

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[peterdsharpe]

This is the right fix, but don't merge this until torchRL pins a torch>=2.5