cdc: core changefeeds don't check kv.rangefeed.enabled until after the initial scan

[HonoreDB]

Describe the problem

On crdb 21.2, a core changefeed on a table with data will emit all the existing rows before it errors with the expected rangefeeds require the kv.rangefeed.enabled setting. This is okay but non-ideal as it potentially separates the operator from the error.

Jira issue: CRDB-13743

[blathers-crl]

cc @cockroachdb/cdc

[stevendanna]

Huh, I thought I had fixed this in e372d9e. I wonder if the issue here isn't the location of the check, but rather that the relevant cluster setting was marked TenantWritable: https://github.com/cockroachdb/cockroach/blob/master/pkg/kv/kvserver/replica_rangefeed.go#L46. when it really should be TenantReadOnly