Attempt to fix hooks using first keyframe at first step of second sampler

[catboxanon]

Attempts to resolve #6183

This fixes the issue where hooks incorrectly use the first keyframe during the first step of a second sampler pass as described in the original issue's workflow. Workflows that sample through all sigmas are not affected by this change.

Unfortunately, I don't believe this resolves the issue entirely, as it still has to invoke the ModelPatcher at the first step of the second sampler pass when it should not be necessary, and while the results are much closer to being identical than before, it still is ever so slightly off. Regardless, I'm opening this PR for further dissection and maybe as a temporary fix until it is properly resolved.

Result of this PR (compare to result 1 in the aforementioned issue)

[Kosinkadink]

The proper 'fix' for guarantee_steps would require that the keyframe logic knows about all the sigmas that the current sampler will experience. Technically guarantee_steps=0 by default would make this problem go away, but then introduces not respecting the start strength when more keyframes exist than steps.

I should be able to make guarantee_steps work as intended with sampling split between multiple KSamplers if I modify the code a bit so the keyframe code knows all the sigmas it will work with. I'll include it with the changes for next ModelPatcher PR I am working on that fleshes out the hooks system. I should have that PR ready by the end of the week, I will refer back to this PR to let you know when it's ready.

[catboxanon]

I should have that PR ready by the end of the week, I will refer back to this PR to let you know when it's ready.

Cheers. I'll choose to leave this PR open for now just for awareness but will close it when that PR is ready.