Use Float8TrainingOpConfig instead of removed FP8GroupedMMConfig alias (pytorch#2573)

## Summary

`FP8GroupedMMConfig` was a temporary backward-compatibility alias in
torchao that has been removed in pytorch/ao#4069. This PR updates
torchtitan to use the canonical `Float8TrainingOpConfig` name directly.

## Change

One-line rename in `torchtitan/components/quantization/float8.py`:
- `FP8GroupedMMConfig` → `Float8TrainingOpConfig` (import + usage)

## Test plan
- No behavior change — `FP8GroupedMMConfig` was an alias for
`Float8TrainingOpConfig` with identical defaults.
- Existing MoE FP8 training tests cover this code path.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-authored-by: Li <lizli102@ctr2-alola-ctrl-01.amd.com>
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

Author: 3 people

torchtitan/components/quantization/float8.py

@@ -274,7 +274,7 @@ def convert(self, model: nn.Module):
         from torchao.quantization.quant_api import quantize_
 
         try:
-            from torchao.prototype.moe_training.config import FP8GroupedMMConfig
+            from torchao.prototype.moe_training.config import Float8TrainingOpConfig
         except ImportError as e:
             raise ImportError(
                 "torchao installation does not have MoE training support. Please install torchao nightly build."
@@ -293,7 +293,7 @@ def moe_module_filter_fn(mod: nn.Module, cur_fqn: str) -> bool:
             model, ["_init_mean", "_init_std"], nn_module_cls=nn.Linear
         )
 
-        config = FP8GroupedMMConfig()
+        config = Float8TrainingOpConfig()
         quantize_(model, config=config, filter_fn=moe_module_filter_fn)
 
         # Re-inject Linear protocol and re-attach attrs