[API Compatibility No.155] Keep input/target alias for bce_with_logits-part

[Aidenwu0209]

PR Category

User Experience

PR Types

New features

Description

• Keep PyTorch-style keyword args for torch.nn.functional.binary_cross_entropy_with_logits during conversion.
• Remove kwargs remap (input->logit, target->label) in PaConvert mapping so converted Paddle code can directly use input/target aliases.
• Add conversion assertion test cases to ensure converted code keeps input/target keywords (including reordered kwargs).
• Related issue: 【启航计划】PaddlePaddle API兼容性增强 Paddle#76301 (No.155)
• Depends on Paddle PR: [API Compatibility No.155] Add kwargs alias for binary_cross_entropy_with_logits-part Paddle#78140

[paddle-bot]

Thanks for your contribution!

[Aidenwu0209]

已定位本次失败原因：当前 CI 使用的 nightly paddle 还未包含 No.155 在 Paddle 侧的 alias 改动（Paddle PR: PaddlePaddle/Paddle#78140 目前仍为 OPEN）。\n\n在本地同版本环境可复现：paddle.nn.functional.binary_cross_entropy_with_logits(input=..., target=...) 报 TypeError: unexpected keyword argument 'input'，因此触发 PR-CI-GPU-UnitTest 失败。\n\n按 #76301 说明，这类 PaConvert PR 需要等 Paddle PR 合入并经过 nightly 同步（通常次日）后再重跑 CI。Paddle 合入后我会第一时间重新触发本 PR CI。