Consider fetching remote env in a different way

[Razz4780]

Currently the agent reads proc filesystem to read env of the target. I imagine an alternative solution, where the env would be taken from Pod template and image

Pros:

• Would allow to drop CAP_PTRACE from the agent.
• User application would see the same env as the targeted container at startup. I think it would be better/feel more natural. Consider flows where env is used as key-value store shared between processes (maybe a bit exotic, but we do this ourselves when doing env request from inside layer)

[aviramha]

This is already being done - today, the env is a mix of the proc file and the pod template. The reason behind that is we were too afraid to break it, so we just merged both instead of replacing. Nowaday I know some users have entry points that add/change env and it might be useful, also some env is provided by k8s itself on start (see all the "SERVICE" env names - not in spec but do exist in the container itself).