Optimize archiver initialization during node restart

[nazar-pc]

Reading blocks in Substrate is slow, like very slow. When we have thousands of blocks to read it could take seconds.

We read blocks from the tip backwards and then forward again when restarting node to restore internal archiver state.

We should be able to accelerate it in two ways:

• use segment headers store to find interested block faster rather than going from tip towards genesis (find_last_archived_block function)
• try reading blocks in parallel, this should help since currently search is CPU-bound and SSDs can easily do more parallel work

[mathematicw]

use segment headers store to find interested block faster rather than going from tip towards genesis (find_last_archived_block function) should be faster and more resource economic