[refactor]Scheduler.set_begin_index

[yiyixuxu]

This refactor affects 16 schedulers that implement step_index counter, except the dpm multistep inverse scheduler (I did not update it; I removed the #copied from statement)

1. added begin_index property and set_begin_index() method to all schedulers that implement the step_index counter. This will allow pipelines to set the first index for the scheduler's step_index counter before the denoising loop. This is particularly useful for img2img pipelines because it may have duplicated first timesteps (see more on https://github.com/huggingface/diffusers/pull/5746/files)
2. added an index_for_timestep method, which contains the logic to convert timestep into sigma index. This is now used by both _init_step_index() and add_noise() methods to decide the very first step_index when begin_index has not been set from the pipeline. The DPM schedulers (dpm-multistep, dpm-singlastep, deis, unipc, sasolver) has slightly different index_for_timestep method from the rest of the schedulers.
3. refactored _init_step_index() and add_noise() methods across the schedulers so they have consistent definitions and added the #Copied from statement whenever possible
4. remove the legacy _index_counter in heun and a couple other schedulers - it was introduced before we had the step_index counter and no longer needed

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[patrickvonplaten]

Suggested change

	@begin_index.setter
	def begin_index(self, index):
	self._begin_index = index
	@begin_index.setter
	def begin_index(self, index):
	self._begin_index = index

Are we using this function?

I don't really see a self.begin_index = ... anywhere

[yiyixuxu]

ok will remove :)

[patrickvonplaten]

Generally this PR looks good to me - just wondering about the self.begin_index setter method, it doesn't seem like we need it.