[schedulers] fix RecursionError in CosineDPMSolverMultistepScheduler

[zxuhan]

What does this PR do?

Fixes #13274.

CosineDPMSolverMultistepScheduler.step constructs BrownianTreeNoiseSampler with the config bounds (sigma_min, sigma_max), but the sampler is then queried with the discretized schedule in self.sigmas, which drifts outside those bounds at both ends:

• The Karras/exponential reconstruction of the endpoints in fp32 lands a few ULPs off. For example with sigma_max=500 on the exponential schedule, self.sigmas[0] is 500.00006103515625. The first query violates torchsde's tb <= t1 check.
• With the default final_sigmas_type="zero" the last query has sigma_next == 0, which is strictly below config.sigma_min. That violates ta >= t0.

torchsde reacts to either by recursively splitting its backing interval until Python's recursion limit is hit, producing the RecursionError: maximum recursion depth exceeded reported for stabilityai/stable-audio-open-1.0 in #13274.

This PR switches the Brownian bounds to the actual extrema of self.sigmas, matching the pattern already used in scheduling_dpmsolver_sde.py. After the change the sampler is always constructed over an interval that encloses every query issued during sampling.

A new regression test in tests/schedulers/test_scheduler_cosine_dpmsolver_multistep.py exercises both Karras and exponential schedules with final_sigmas_type="zero". Without the fix the exponential subtest reproduces the RecursionError; with the fix both subtests pass.

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? See [Bug][AMD] stabilityai/stable-audio-open-1.0 encounter RecursionError: maximum recursion depth exceeded #13274.
• Did you make sure to update the documentation with your changes? No user-facing documentation needs an update; the fix only changes scheduler internals.
• Did you write any new necessary tests?

Who can review?

@yiyixuxu @sayakpaul