Nodedb followup: SafeNoSync

[AskAlexSharov]

Follow-up after: #13457 (comment)

Set mdbx_env_set_syncperiod and mdbx_env_set_syncbytes and call in goroutine mdbx_env_sync_ex.

mdbx doesn't have own thread: at all.
In SafeNoSync - it can call fsync in 2 cases:

• or inside Commit (if some of limits reached in this rwtx)
• or inside mdbx_env_sync_ex

So, SafeNoSync - increase latency of "unlucky" rwtx which reached limits. But we can reduce this probability by manually creating goroutine with periodical call of mdbx_env_sync_ex (then this goroutine will be blocked instead of "unlucky Commit")

I prefer SafeNoSync over .Batch because:

it allow reduce latency for most of writers. it doesn't hold writers until fsync. (maybe 1 writer holded - to perform fsync by this thread. i'm not 100% sure - maybe mdbx has background thread for fsync).
so with .Batch if set DefaultMaxBatchDelay to 2sec it will add latency of 2sec to all writers. but in SafeNoSync if set mdbx_env_set_syncperiod to 2sec it will not add any latency to writers.

with SafeNoSync - if kill -9 came in wrong time - it may leave DB in "bad" state:

• if then open DB in rw mode: then mdbx can self-recover from "bad" state.
• but if open DB in ro mode: then mdbx can't recover and exit with error (can't open db).

Let's remove ReadOnly mode of DB. RPCDaemon and integration tools - nothing will change if open in RW mode.

[antonis19]

Closing with: #13980