fix: optimize transitive code search tool

[zvigrinberg]

Context:

Transitive code search tool returns tuple of 2, the first returned value in the tuple is a boolean indicates whether the vulnerable function is called/reachable from the inspected app, and the 2nd returned value is a path of that call, if the former it True. the path comes in the form of a list of documents.(each document object represent a function/method in the programming language).

Currently, the LLM doesn't use the content of the returned functions documents in the path for subsequent llm calls or reasoning, and it doesn't bring any added value, on the contrary, it's just loading and bombing the context window size, in vain ( and it causes from time to time the LLM Request to crash on Bad request 400, because of context window Exceeding).

Solution :

instead of full function definition per each function in the path, just put the function name alone to make the output of the tool effective in the same way, but much more compact to save and reduce a lot the LLM requests' context size and minimize as possible the total Token count for the heavy agent loop LLM Node.

[zvigrinberg]

/test-heavy