Rentention strategies don't work as expected

[kalabukdima]

The most common scenario for us is:

• Set the initial from_head(K) to start syncing even if there is no archival data
• Every time there is a new index of archival chunks, call retain with from_block set to the current archive height to clean up the old blocks.

However, it fails with heavy consequences in the following scenario:

• Archives stop syncing, so we periodically call retain with from_block with the same block N
• When Head - K becomes greater than N the storage starts to clean blocks up to Head - K
• The next time we call retain(N) it says that there is a gap and drops the entire database
• It then starts ingestion from some block above the finalized head (2 blocks from head)
• This repeats on every consequent call to retain(N)

I'd say the database should never be fully dropped without manual intervention. Once we run hotblocks in a separate service, we can add an internal endpoint for that.