What does this PR do?

Fixes duplicate timesteps issue in DPMSolverMultistepScheduler when using various sigma conversion methods (Karras, Lu, exponential, beta) combined with beta_schedule="squaredcos_cap_v2".

Before this PR

• Timesteps were not being rounded for squaredcos_cap_v2 schedule
• This caused multiple sigma values to map to the same timestep
• Example: timestep 998 appeared 9 times instead of once
• Resulted in "index out of bounds" errors during sampling

After this PR

• All sigma conversion methods now properly round timesteps
• Duplicate timesteps are detected and removed
• Users are warned when fewer unique timesteps are generated
• Sampling completes without errors

Changes Made

1. Added _ensure_unique_timesteps() helper method to handle duplicate timesteps consistently
2. Updated all sigma conversion methods to unconditionally round timesteps
3. Applied duplicate detection and removal to:
   • use_karras_sigmas
   • use_lu_lambdas
   • use_exponential_sigmas
   • use_beta_sigmas
4. Added comprehensive test test_no_duplicate_timesteps_with_sigma_methods

Testing

✅ All 42 existing tests pass
✅ New test passes
✅ No regression detected
✅ Tested with multiple sigma methods and beta schedules

Fixes #12771

Before submitting

• Did you read the [contributor guideline]?
• Was this discussed/approved via a GitHub issue?
  DPMSolverMultistepScheduler out-of-bounds error when using beta_schedule="squaredcos_cap_v2" + use_karras_sigmas=True #12771
• Did you write any new necessary tests?

Who can review?

@yiyixuxu