Keep main's local-tensor marker and let the quantizer replicate scalars
The rename of `_hf_quantized_needs_local_tp` had no user here and collides with #49679. Scalar replication
goes back from `resolve_parallel_plans` into `HfQuantizer.update_parallel_plan`, as Arthur asked on #49015,
so every quantizer gets it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>