feat(config): add fused and foreach optimizer controls

[CharlesXu-HQ]

Description

Addresses the remaining optimizer-control request in #273.

Motivation

PyTorch can select the fused and foreach optimizer implementations automatically. Explicit control is useful when tuning optimizer memory usage or working around backend-specific behavior.

Changes

• Add optional fused and foreach fields to OptimizerConfig.
• Forward explicitly configured values to veRL's FSDP/FSDP2 override_optimizer_config.
• Preserve backend defaults when the options are unset.
• Keep Megatron and critic optimizer behavior unchanged.
• Document both options in the English and Chinese configuration guides.

Testing

• Targeted optimizer tests: 3 passed, 4 subtests passed.
• Configuration suite on Python 3.12 / RTX 5090: 18 passed, 1 multi-node Ray test deselected, 4 subtests passed.
• Pre-commit checks on all changed files passed.
• CUDA AdamW smoke test with fused=True, foreach=False passed.

Checklist

Please check the following items before code is ready to be reviewed.

• Code has passed all tests
• Docstrings have been added/updated in Google Style (not applicable; no public callable added)
• Documentation has been updated
• Code is ready for review