`pg_autoctl show state --local` should be read-only

[tboevil]

pg_autoctl show state --local will be executed keeper_init，then it will delete pg_autoctl.pg. external components calls to this command may cause conflicts with the keeper.

T0+0.13s  openat(O_CREAT) pg_autoctl.pg by `pg_autoctl: node active`
T0+0.18s  unlink pg_autoctl.pg by `pg_autoctl show state --json --local`
T0+9.4s   keeper write again
T0+10.2s  unlink again
...       30s unlink retry(ENOENT after first attempt)
T0+30s    keeper timeout → pg_basebackup

[dimitri]

Thanks for the clear strace-level trace -- it made this one easy to pin down precisely.

Fixed by #1175 (merged as e1d6b3f). show state --local went through the same keeper_init() path as every state-changing command, which unconditionally deletes pg_autoctl.pg -- the postgres controller's only signal for whether Postgres should be running or stopped. That's correct for commands actually taking over Postgres management, but --local's whole point is to be a read-only peek at local state, so it had no business doing that.

Added keeper_init_read_only() / local_postgres_init_read_only(), narrowly scoped to skip that deletion, and switched --local to use them. Every other caller of keeper_init() (enable/disable maintenance, node drop, etc.) is untouched -- those are legitimate state-changing operations where clearing a stale status file on entry is still correct.