delete_retention_ms migration is not working when included in redpanda.yaml

[mattschumpert]

Version & Environment

Upgrading to 23.3 via apt update_upgrade on Ubuntu 22

1. We changed delete_retention_ms to log_retention_ms to be in line with Kafka, but we were supposed to also do a smooth migration of the config value on upgrade to transition the cluster smoothy to the new property. It appears we may not have done this as an upgraded cluster rejects the config in redpanda.yaml (or maybe we migrated it in centralized configuration but can't accept the config from a redpanda.yaml with a hard failure.

2. We're also not supposed to even remove the legacy config (which is supposed to be aliased) until 24.1.

What is the expected behavior here when sunsetting this config. I guess at some point we can stop accepting it from an external YAML, but at least that is not supposed to happen until 24.1. At most, this should be a deprecated config WARNING that allows Redpanda to start.

We should also double check that we're migrating the centralized cluster config properly from delete_retention_ms to log_retention_ms as planned in our metadata,

Report from beta user:

I had to comment out this line in our Redpanda.yaml, probably a leftover old config that’s in the wrong spot.

redpanda:

        delete_retention_ms: 7776000000

Error was “Failure during startup: std::invalid_argument (Unknown property delete_retention_ms)”

[andijcr]

i'm pushing some commits to fix this here #15605 , need to work out a test for this