schedulers/scheduling_dpmsolver_multistep: round timesteps for exponential and beta sigma schedules

[Dev-X25874]

What does this PR do?

In DPMSolverMultistepScheduler.set_timesteps, the use_karras_sigmas and
use_lu_lambdas branches both call timesteps.round() after computing
timesteps via _sigma_to_t (which returns floating-point values through
interpolation), before those timesteps are cast to torch.int64.

The use_exponential_sigmas and use_beta_sigmas branches were missing this
.round() call, causing raw float values to be truncated (not rounded) on
the torch.int64 cast at the end of set_timesteps. This silently introduces
a systematic off-by-one error in the noise schedule index lookup whenever
either of these two sigma modes is used with this scheduler.

This PR adds timesteps = timesteps.round() to both branches, consistent with
the existing pattern already established by use_karras_sigmas and
use_lu_lambdas.

Fixes # (issue)

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline?
• Did you read our philosophy doc (important for complex PRs)?
• Was this discussed/approved via a GitHub issue or the forum? Please add a link to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

Who can review?

@DN6

[Dev-X25874]

Hey @DN6 — could you please take a look at this PR when you get a chance? It fixes a silent off-by-one error in DPMSolverMultistepScheduler.set_timesteps where the use_exponential_sigmas and use_beta_sigmas branches were missing a .round() call before the torch.int64 cast — causing float truncation instead of rounding. The use_karras_sigmas and use_lu_lambdas branches already do this correctly; this PR just brings the other two in line. Two-line fix, no behaviour change for any other code path. Thanks!