Replace deprecated torch.range in hy3dgen/texgen/hunyuanpaint/pipeline.py with torch.arange

Summary

hy3dgen/texgen/hunyuanpaint/pipeline.py:598 calls torch.range, which is deprecated and emits a UserWarning on every call ("torch.range is deprecated and will be removed in a future release because its behavior is inconsistent with Python's range builtin. Instead, use torch.arange, which produces values in [start, end)." — verified on torch 2.11.0). It is the only torch.range usage in this repo:

if self.is_turbo:
    bsz = 3
    N_gen = 15
    index = torch.range(29, 0, -bsz, device='cuda').long()   # ← emits UserWarning on every call
    timesteps = self.solver.ddim_timesteps[index]

This runs in the turbo path of the paint pipeline on every generation.

Proposed fix

torch.range is end-inclusive: torch.range(29, 0, -3) produces [29, 26, 23, 20, 17, 14, 11, 8, 5, 2] (10 values; 0 is not reachable with step -3). The torch.arange equivalent (open interval, step excluded end) is:

index = torch.arange(29, -1, -bsz, device='cuda').long()

(-1 = last included value 2 + step -3; torch.arange with integer inputs returns int64, so the explicit .long() is a no-op — kept to match the surrounding style.)

Validation (torch 2.11.0+cpu)

• torch.equal(torch.range(29, 0, -3), torch.arange(29, -1, -3)) → True (identical values, end-inclusive semantics preserved).
• With warnings.simplefilter("error", UserWarning): the current line raises the deprecation UserWarning; torch.arange constructs cleanly.

Impact

• No deprecation UserWarning during turbo inference on any supported torch version; identical timestep indices.

References

• Issue #270: "Hunyuan3D-2 Warnings and Issue Resolution Methods" — user already reported this exact torch.range warning (2025-06-09) and patched it locally; no maintainer response, code still unfixed.
• PyTorch docs: torch.arange — recommended replacement.
• PyTorch docs: torch.range (deprecated) — official deprecation note ("Instead, use torch.arange(), which produces values in [start, end).").