Drop the legacy FP8 copies of the plan rewrites and test the new mechanisms
The legacy FP8 quantizer replicated 0-dim params and swapped the Mega MoE styles itself.
`resolve_parallel_plans` does both, so its override and `FP8Experts._impl_tp_layer_overrides` go.
Make the plan helpers private and in place, inline the local-tensors marker check, and trim the comments.
Test the intra-expert placements, the plan resolution and a non-switchable implementation.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>