Prompt embeds update for T2I, I2I and upscale.

[JJassonn69]

• the all the models worked nicely except timbrooks/pix2pix, when prompting use + or - after the word to increase the weight of the word in generation, you can add multiple ++ or -- to increase or decrease weight.

• the inferencing time didnot change at all most of the time or was barely reduced for some cases.

• example of prompt "a cat playing with a ball++" , here the ball has higher weight so it will be more focused when generating.

[JJassonn69]

@ad-astra-video , i have cleaned up and rebased it using the main branch from ai-worker. In regards to the changing of embeddings from Compel to another is out of the scope of the bounty initially listed. We might have to add another bounty for that as it will require extensive testing with all the models and pipelines. initial bounty > Add pipeline support for prompt Embeds