Storing only metadata

[rayruizhiliao]

This PR addresses

• storage_objects contains full JSON lines #31

[dimavrem22]

since the purpose of this step is to identify potential sources of tokens, api keys, etc., i think there might be a more effective approach. we can...

1. try to resolve the path to the searched substrings ourselves and provide those paths to the llm as context (basically telling it the answers); this way the llm just has to 'verify' that the paths are correct
2. dump all of the transactions and session storage related items into a vectorstore and force the agent to query the vector store for these new files before returning a response to us
3. [most complex version] do 1. and also give the llm explicit tools to parse the dicts we have (transaction response body or the session storage value)