[float8 moe] Support TP (#1375)

We can remove this assertion, TP support for float8 rowwise MoE training
was added in this PR stack: pytorch/ao#2473

Author: danielvegamyhre

torchtitan/components/quantization/float8.py

@@ -59,9 +59,6 @@ def __init__(self, job_config: JobConfig, parallel_dims: ParallelDims):
 
         # Validate MoE training prototype limitations.
         if self.moe_fqns:
-            assert (
-                job_config.parallelism.tensor_parallel_degree == 1
-            ), "Float8 MoE training prototype does not yet support tensor parallelism"
             assert (
                 job_config.parallelism.pipeline_parallel_degree == 1
             ), "Float8 MoE training prototype does not yet support pipeline parallelism"