[TorchFX][Optimization] Run eliminate dead code once for constant correction commands

🚀 Feature request

https://github.com/openvinotoolkit/nncf/pull/2882/files#r1735077003

• Look like eliminate_dead_code should be run ones after apply all bias_update transormation.
  Or maybe will be better to use something like self.erase_node(old_constant_node) to remove only old constant node to avoid loops for all nodes.
  As i understand eliminate_dead_code here removes only old constant node. And it's actual for all usage eliminate_dead_code in transformation.py.

• We can do it, but we should give up on transformations and move to commands for that. Leaf insertion transformation not expecting eliminate_dead_code call yet model transformer could not split bias correction and leaf insertion commands apart

Feature Use Case

No response

Are you going to submit a PR?

• Yes I'd like to help by submitting a PR!

Start in transformation.py and trace the TorchFX bias_update transformations and their uses of eliminate_dead_code. Compare the proposed single cleanup after all bias corrections with removing the old constant node or moving to commands; done means the selected approach handles constant correction without unnecessary repeated cleanup or missed node removal.