Deterministic generation

[drakfyre]

Added deterministic generation (and it works this time...)

So now you can re-generate pictures that were made with CADS! :>

[v0xie]

Hey thanks so much for this, reproducibility is a really issue this repo has yet to tackle.

When testing this PR, I'm running into an error in the latest commit:

** Error executing callback cfg_denoiser_callback for F:\stablediffusion\stable-diffusion-webui\extensions\sd-cads-cfg\scripts\cads.py
    Traceback (most recent call last):
      File "F:\stablediffusion\stable-diffusion-webui\modules\script_callbacks.py", line 216, in cfg_denoiser_callback
        c.callback(params)
      File "F:\stablediffusion\stable-diffusion-webui\extensions\sd-cads-cfg\scripts\cads.py", line 112, in <lambda>
        y = lambda params: self.on_cfg_denoiser_callback(params, t1=t1, t2=t2, noise_scale=noise_scale, mixing_factor=mixing_factor, rescale=rescale, total_sampling_steps=total_sampling_steps)
      File "F:\stablediffusion\stable-diffusion-webui\extensions\sd-cads-cfg\scripts\cads.py", line 160, in on_cfg_denoiser_callback
        rng.manual_seed(params.seed)
    AttributeError: 'CFGDenoiserParams' object has no attribute 'seed'

Previously I took a stab (unsuccessfully) at reproducing seeds in this test branch: https://github.com/v0xie/sd-webui-cads/tree/fix-rng

I was kind of worried setting the manual seed would alter existing seeds (i.e. the image that would be generated without any effect from CADS), so I tried using A111 rng.randn_local to generate the noise at each iteration, but that didn't work for me. If it's as easy as setting the manual seed I'm all for it.