[BUG] SliceSampler should return unique IDs when sampling multiple times from the same trajectory

Describe the bug

When using SliceSampler, with strict_length=False, the documentation recommends the use of split_trajectories. However, if two samples from the same episode are placed next to each other, this produces the wrong output because subsequent samples may have the same trajectory_key despite being logically independent.

To Reproduce

import torch
from tensordict import TensorDict
from torchrl.collectors.utils import split_trajectories
from torchrl.data import ReplayBuffer, LazyTensorStorage, SliceSampler

rb = ReplayBuffer(storage=LazyTensorStorage(max_size=1000),
                  sampler=SliceSampler(
                      slice_len=5, traj_key="episode",strict_length=False
                  ))

ep_1 = TensorDict(
    {"obs": torch.arange(100),
    "episode": torch.zeros(100),},
    batch_size=[100]
)
ep_2 = TensorDict(
    {"obs": torch.arange(4),
    "episode": torch.ones(4),},
    batch_size=[4]
)
rb.extend(ep_1)
rb.extend(ep_2)

s = rb.sample(50)
t = split_trajectories(s, trajectory_key="episode")

split_trajectories returns nonsense results when trajectory_key contains non-contiguous duplicates.
Even if that weren't the case, there would still be a bug:

When SliceSampler is drawing from relatively few trajectories, there will be situations where multiple slices of the same trajectory are returned next to each other:

episode 0  0  0  0  0  0  0  0  0  0...
obs     2  3  4  5  6  41 42 43 44 45...
        |-1st slice-|  |-2nd slice--|

However, split_trajectories will see that episode is the same for both slices, and incorrectly combine them into one longer slice.

Expected behavior

SliceSampler should add an additional key to its returned dict to distinguish samples, at least when strict_length=False:

episode 0  0  0  0  0  0  0  0  0  0...
obs     2  3  4  5  6  41 42 43 44 45...
slice   0  0  0  0  0  1  1  1  1  1

Screenshots

If applicable, add screenshots to help explain your problem.

System info

M1 Mac, version 15.1

import torchrl, numpy, sys
print(torchrl.__version__, numpy.__version__, sys.version, sys.platform)

0.6.0+7bf320c 1.26.4 3.11.9 (main, Apr 19 2024, 11:44:45) [Clang 14.0.6 ] darwin

Both torchrl and tensordict were installed from source.

Checklist

• I have checked that there is no similar issue in the repo (required)
• I have read the documentation (required)
• I have provided a minimal working example to reproduce the bug (required)