EPP upgrade/downgrade guide

[liu-cong]

What would you like to be added:

We should provide guidance on upgrading/downgrading EPP versions to minimize downtime. Depending on the tradeoff between risk and speed, I imagine there can be different upgrade strategies:

1. Simply do a rolling update of the EPP deployment. With the active-passive HA mode EPP HA deployment #692, a minimal disruption to traffic may be expected when switching the leader.
2. Blue-green upgrade. Deploy a new EPP deployment, verify it works, and update the ExtensionRef.
3. Create a new InferencePool (still pointing to the same model servers) with new EPP, and configure HTTPRoutes to split traffic between new and old inference pools.

Why is this needed:

[danehans]

xref #735 as upgrade/downgrade should be part of the ops guide.

[liu-cong]

We don't currently have an owner for this issue. I suggest we move this to v0.5 release to not block the upcoming v0.4 release.

[ahg-g]

sounds good to me

[nirrozenbaum]

++