[mps] Fix float64 upcast in HeliosDMDScheduler.convert_flow_pred_to_x0

[Shreyas-jk]

What does this PR do?

HeliosDMDScheduler.convert_flow_pred_to_x0 upcasts its tensors to float64 via
.double() for higher-precision math. The MPS backend has no float64, so this
raises Cannot convert a MPS Tensor to float64 dtype as the MPS framework doesn't support float64 on Apple Silicon the moment step() calls it.

The fix mirrors the existing # mps does not support float64 pattern already
used across the schedulers (e.g. scheduling_flow_match_euler_discrete.py):
compute in float32 on mps, keep float64 on CUDA/CPU. Off-MPS numerics are
unchanged. Same fix shape as #13701.

Testing

Verified locally on Apple Silicon (MPS available):

• New regression test in tests/schedulers/test_scheduler_helios_dmd.py fails
  on the unpatched method with the float64 TypeError, and passes with the fix
  (red → green).
• make style && make quality clean.
• pytest tests/schedulers/ -k "flow or helios" → 16 passed.

The test is gated with @unittest.skipUnless(torch_device == "mps", ...), so it
skips on the Linux/CUDA CI runners and runs for contributors on Apple Silicon.

cc @yiyixuxu @sayakpaul

[github-actions]

Hi @Shreyas-jk, thanks for the PR! It does not appear to link an issue it fixes. If this PR addresses an existing issue, please add a closing keyword (e.g. Fixes #1234) to the PR description so the issue is linked. See the contribution guide for more details. If this PR intentionally does not fix a tracked issue, a maintainer can add the no-issue-needed label to silence this reminder.