Update DDP strategy to handle unused parameters consistently

[dohyeonYoon]

Removed segmentation head condition for DDP strategy to consistently apply unused parameters detection across detection and segmentation models.

What does this PR do?

**Related Issue(s):Related to #1093

Type of Change

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

Testing

• I have tested this change locally
• I have added/updated tests for this change

Test details:

I used the code below, and the error is gone.

import os
from rfdetr import RFDETRMedium

model = RFDETRMedium()

model.train(
    dataset_dir="/home/dohyeon/Datasets/dataset_rfdetr_576",
    epochs=1,
    batch_size=4,
    grad_accum_steps=2,  # 2GPU × grad_accum_steps=2 × batch_size=4 = effective 16
    output_dir="output/strategy1",
    progress_bar="tqdm",
    devices=2,
    # strategy="ddp_find_unused_parameters_true",  # NOT set → triggers error
    num_workers=16,
    pin_memory=True,
    resolution=576,
    resume=None,
    checkpoint_interval=1,
    use_ema=True,
    tensorboard=True,
    seed=42,
)

Checklist

• My code follows the style guidelines of this project
• I have performed a self-review of my own code
• I have commented my code where necessary, particularly in hard-to-understand areas
• My changes generate no new warnings or errors
• I have updated the documentation accordingly (if applicable)

Additional Context

This bug only occurs when training with multiple GPUs, for example using:

torchrun --nproc_per_node=2 train.py

The error occurs only when using two or more GPUs. It does not occur when running on a single GPU:

python train.py

[CLAassistant]

All committers have signed the CLA.