Infer intent made into it's own function and isVideo, isAccessBlocked determination added to it as well

[sarge1989]

Issue

• Resolves isVideo and isAccessBlocked being flagged even though it's not the initial message sent in #59

Description

• Instead of letting infer_intent be a tool that the agent calls, enforced a fixed starting step (preprocess_inputs)
  • Extract URLs from the text sent in if it's a text message and take screenshots of the URLs
  • Get LLM to 1) determine isVideo 2) determine isAccessBlocked 3) infer intent
• Agent prompt now adjusted not to infer intent, but instead accept intent as input (from first step)