Fixes: #2417

• tokio::task-s and tokio blocking tasks (spawn_blocking) are now spawned and tracked via a task tracker, using the util::task::spawn[_blocking] api,
• cancellation is triggered via cancellation token,
• the token is also passed to spawn_blocking, and the caller must use it to bail out in case of a longer running task,
• std::thread-s should now be spawned with util::task::spawn_std, because it gives the caller access to the token too and the caller should then use it to break out of a long running loop,
• after cancellation is triggered, util::task::spawn[_blocking] tasks will be waited upon, while util::task::spawn_std will not be joined, unless the caller locally does so (we don't use many and I don't think their detachment is a big issue),
• INT and TERM signal listeners are installed after the db is migrated and tries are pruned to allow the user to cancel that process instead of forcing them to wait,
• from then on the process does not exit, even when an error is encountered, which is handled in the final select - this is to ensure orderly cancellation even in case of startup errors

Apart from implementing the above ⬆️ , adds a refactor plus some fixes to bugs noticed on the way, as per commit order:

• add a util crate for general Rust utilities,
• use rayon where spawn_blocking is sub-optimal,
• move make_stream and AnyhowExt to util,
• rollback_to_anchor does not commit() the db transaction,
• remove some dead code,
• p2p db connection pool is too small,
• critical task errors not propagated to process return status,
• make sure that all db connection pools, especially those that are RO are dropped before the last RW connection pool is dropped (if RO connection pool is dropped last, wal and shm files are never cleaned up).