schedulers/flow_match_euler_discrete: fix double-shift bug by storing sigma_min/sigma_max before shift is applied

[Dev-X25874]

What does this PR do?

Fixes #13243

FlowMatchEulerDiscreteScheduler.__init__ was storing sigma_min and sigma_max after the shift formula had already been applied to sigmas:

sigmas = shift * sigmas / (1 + (shift - 1) * sigmas)  # shift applied here
...
self.sigma_min = self.sigmas[-1].item()  # already-shifted value stored
self.sigma_max = self.sigmas[0].item()  # already-shifted value stored

Then in set_timesteps, when no custom sigmas/timesteps are provided, the scheduler reconstructs a linspace from sigma_max → sigma_min (the shifted values), divides back by num_train_timesteps, and applies the shift formula a second time — resulting in a doubly-shifted sigma schedule that diverges from what the model was trained with.

This PR moves the sigma_min/sigma_max assignment to before the shift is applied, so set_timesteps always starts from the correct unshifted linear range and applies the shift exactly once.

The fix is surgical: self.sigmas and self.timesteps stored in __init__ are unchanged. Only the two boundary reference scalars are corrected. Users passing custom sigmas or timesteps directly to set_timesteps are entirely unaffected.

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?

@yiyixuxu