Add support of block header pruning.

[shamil-gadelshin]

This PR follows #3962 and relates to #1570

The PR introduces --prune-block-headers flag to CLI that enables block header pruning along with block prunings when conditions (pruning mode, block finalization, etc) are met. The rationale behind the change is to reduce DB size by removing unnecessary data from the disk.

The flag goes to prune_blocks and prune_displaced_branches methods of the sc-client-db crate and enables the gathering of the affected hashes to prune, the hash collection is iterated with block header pruning (both from DB and metadata cache) in try_commit_operation as one of the final steps before the transaction commits. The initial implementation was simpler and pruned the block headers together with block pruning, however, try_commit_operation refreshes the cache after the block header pruning and before the transaction commit and thus keeps the block header in memory.

Tests cover regular block header pruning and block header pruning on forks and reorgs.

[shamil-gadelshin]

Related forum post: https://forum.polkadot.network/t/block-header-pruning/7198