Skip installing large aliased constant-folded weights

[micwill755]

Summary

• Skip installing large constant-folded tensors that only alias existing module weights, avoiding expensive cpu().contiguous() copies under offload_module_to_cpu.

This is PR 2/3 in a stack. It includes the partition short-circuit PR plus this commit. Review only the latest commit, or this compare:
micwill755/TensorRT@partition-short-circuit...skip-aliased-fold-install

Test plan

• Flux NVFP4 cold compile still succeeds with offload_module_to_cpu=True
• Large weight permutes remain in-graph (not installed as frozen params)
• VLA still folds materialized embedding constants and compiles (int32 indices)

[micwill755]

Superseded by the real GitHub stack (stack #4533): #4530 -> #4531 -> #4532. These fork branches could not be stacked because GitHub stacked PRs do not support pull requests from forks.