Replace individual input tools with a WebDriver-like API

[OrKoN]

Is your feature request related to a problem? Please describe.

Currently, we have hover, click and press_key tools which are rather low level and to send multiple of them the client needs to be able to issue multiple tool calls (which might involve multiple LLM round trips). In addition, we support some low level input commands but not all like the perform_actions API.

Describe the solution you'd like

WebDriver defines the perform actions API https://www.w3.org/TR/webdriver-bidi/#command-input-performActions that allows sending a sequence of actions in one tool call. We can replace our existing low level APIs with an API similar to WebDriver (but integrated with our other concepts like uids). We should not replace higher level APIs though, like fill, fill_form, drag.

Describe alternatives you've considered

Current approach.

Additional context

We need to test if the models have any difficulties using the WebDriver like API.

[OrKoN]

cc @natorion

[natorion]

I think this would be the right move, depending on the impact on efficacy of the tools of course.

Should this happen before the CLI wrapper?

[OrKoN]

I think it is not strictly dependent on each other but individual input commands might be easier for the CLI.