Fix timesteps of DPMSolverMultistepScheduler. (#88)

There were minor differences in the timesteps because the linspace was
computed slightly differently. This PR makes the Swift implementation
identical to the current Python implementation in diffusers, which was
originally contributed by the DPM-Solver++ author.

See
https://github.com/huggingface/diffusers/blob/main/src/diffusers/schedulers/scheduling_dpmsolver_multistep.py#L199-L204
for reference.

Author: pcuenca

swift/StableDiffusion/pipeline/DPMSolverMultistepScheduler.swift

@@ -77,7 +77,7 @@ public final class DPMSolverMultistepScheduler: Scheduler {
         self.sigma_t = vForce.sqrt(vDSP.subtract([Float](repeating: 1, count: self.alphasCumProd.count), self.alphasCumProd))
         self.lambda_t = zip(self.alpha_t, self.sigma_t).map { α, σ in log(α) - log(σ) }
 
-        self.timeSteps = linspace(0, Float(self.trainStepCount-1), stepCount).reversed().map { Int(round($0)) }
+        self.timeSteps = linspace(0, Float(self.trainStepCount-1), stepCount+1).dropFirst().reversed().map { Int(round($0)) }
     }
 
     /// Convert the model output to the corresponding type the algorithm needs.