Optimize block finalization

[nazar-pc]

First commit simply moves code around to make review easier, second implements workaround described in #3321 (comment), which should improve situation significantly and I think we can consider that it resolves #3321 for now (thought there are still significantly optimization possibilities remaining in Substrate itself).

Did a quick test locally, it should at very least not break thing comparing to what we already have.

Code contributor checklist:

• I have read, understood and followed contributing guide

[nazar-pc]

test_domain_tx_propagate hanged on macOS again

[NingLin-P]

Should it be notification.header.number() >= &importing_block_number? otherwise, there can be a race condition where the block is imported before we call client.every_import_notification_stream().

[nazar-pc]

We call this from the context of block import, so for practical purposes Substrate right mow will only import one block at a time. But even if not, it doesn't really matter because all we really try to do here is delay finalization, it will work fine even if we don't, just causes unnecessary delays in block propagation as described in #3321

[NingLin-P]

Okay, I see, the acknowledgement in the block notification will avoid such race condition happening. Though, my concern is more about the archiver being stuck in this loop... Perhaps instead of block_importing_notification_stream, we should use every_import_notification_stream to derive the archiver forward so we don't need to wait for block import to finish.

[nazar-pc]

We are using every_import_notification_stream though or am I missing something?

[NingLin-P]

I mean to replace this stream with every_import_notification_stream:

subspace/crates/sc-consensus-subspace/src/archiver.rs

Lines 950 to 953 in eed2c70

	// Subscribing synchronously before returning
	let mut block_importing_notification_stream = subspace_link
	.block_importing_notification_stream
	.subscribe();

[nazar-pc]

That is importing, not import stream. It is already fired for every single block before it is imported. We can't replace it with stream that processes blocks after import without breaking archiving.