[Bug]: Donot expect clean output when running bench command

[Bowrna]

Information about bug

When running agent job and when executing the bench command, do not load the data assuming that the data parsed will be a perfect json. Having custom apps in the folder and that could contain print statement could mess up the output of the bench commands. Say example in this case, we are getting installed apps.
Expected output:
bench --site hrms.frappe.cloud execute frappe.get_installed_apps
["frappe","erpnext"]

But for a user with custom apps and having logged prints in std out:
Actual output:
bench --site randomhrms.frappe.cloud execute frappe.get_installed_apps
✅ Duckwalk override: ...
["frappe", "erpnext", "insights", "greendigit"]

We can parse the output and figure out the expected pattern using regex and use that as input to run this job step. Otherwise the jobs will get failed.

https://github.com/frappe/agent/blob/bf88493ed9998c17fc6b42153577ef9f86ce564a/agent/site.py#L632-640

Steps to Reproduce

Run a new site from backup job in bench with custom apps having stdout prints.

In my case, the output for the bench looked like the above one.

Relevant log output / Stack trace / Full Error Message.