[Advanced Paste] Add Semantic Kernel opt-in to allow chaining of paste actions

[drawbyperpetual]

Description of the new feature / enhancement

Add new toggle "Enable advanced AI":

• Off when "Enable Paste with AI" is already enabled (so no change for existing users).
• On when explicitly toggled or when "Enable Paste with AI" is activated (so new users get the feature immediately).

This feature allows for the use of Semantic Kernel to determine a "chain" of paste-actions to be performed when using "Paste with AI." This means Paste with AI can now:

• ...Work with non-text input like images.
• ...Produce non-text output like files.
• ...Execute multiple transformations (in a chain, e.g. Image to text -> Text to JSON text -> JSON text to TXT file).
• ...Produce meaningful AI-generated error messages.

Example queries

• Explicit steps: (with image on the clipboard) "Convert this image to text using OCR, translate the text to French and then save the result as a TXT file"
• Implicit steps: (with image on the clipboard) "Translate to French and save as a TXT file"

Screenshots

Scenario when this would be used?

Semantic Kernel is useful in any situation when multiple transformations are desirable. As the catalog of actions available in Advanced Paste grows, this feature will become more and more useful.

Currently, this feature is especially useful in:

• ...Allowing text transformations to work on images containing text (e.g. "Translate to French" now works on text as well as on images).
• ...Allowing intermediate transformations to first occur when saving data as files (e.g. "Translate to Spanish and save as a TXT file").

Supporting information

Further details:

• Semantic Kernel currently uses OpenAI as its backend, reusing the user's credentials that are used for regular text -> text AI transformations.
• Semantic Kernel uses gpt-4o. Regular text -> text AI transformations continue to use gpt-3.5-turbo-instruct for now.
• Generally speaking, queries using Semantic Kernel consume more OpenAI API credits.
• When Semantic Kernel is used for Custom Actions, the chain of actions executed is cached on disk, meaning subsequent executions of a custom action with the same available clipboard formats should be faster and not consume further OpenAI API credits (unless of course text -> text AI transformations are called as 'sub-routines').