Figure out the mechanism by which a map resize will be made in all cases in lmdb

[TheQuantumPhysicist]

While the current lmdb backend resizes in most cases automatically, there's a chance for MDB_MAP_FULL error to be returned whether in:

• writes
• commits

and we should be able to handle these cases gracefully and not halt the program execution. Currently, in our code, we're not fully invulnerable to that problem. For example, in chainstate, when attempting to write a new block, we just "retry" if we get a recoverable error. However, if we get an MDB_MAP_FULL error, we should make an attempt to resize the database.

Attempts are being made to automate this, but this should be checked in LmdbImpl again later when the node is ready.

[TheQuantumPhysicist]

To close this case, a test for the node should be done, where we see that a database commit failed in process_block, but a retry was made and it worked.

[TheQuantumPhysicist]

There was an error reported once by @pavel-kokolemin that resizing failed and the node crashed. Unfortunately the data directory for that was lost. We should be careful and monitor this and ensure it's done correctly.

EDIT: This issue never seem to have happened again. There's a chance that this is reasonably stable that we don't need this for mainnet.