Generate object mappings incrementally

[teor2345]

This PR generates object mapping when each block is added to the archiver, rather than only at the end of a full segment. When a part of a block added in Archiver::with_initial_state(), those mappings are generated when the next block is added.

This reduces load bursts on the mapping indexer, and reduces object submission to mapping latency.

Piece generation still requires a full segment.

Cleanups

• Simplify segment item access

Code contributor checklist:

• I have read, understood and followed contributing guide

[nazar-pc]

This is exactly the approach I was imagining, thanks!
Mostly have nits about internal APIs and suggested improvements to comments. The only thing I don't understand is the change of with_initial_state() API.

Maybe we could set a 40 minute timeout for the test, instead of the default 6 hours?
I’ve only ever seen them take about 25 minutes.

We use self-hosted runners that are much more powerful than the basic official runners, but our CI is configured to not depend on our runners in any way and to remain fully functional in forks (so contributors can run all the tests, produce custom builds, etc.). Latest run in my fork took ~1h25m, so I think decreasing time limit for tests and clippy to 2h is probably fine, but I wouldn't go as low as 40 minutes.

[teor2345]

I’ll give these changes a shot now.

[teor2345]

I think this might be what you want?

[nazar-pc]

I think produce_object_mappings should process all segments like it did before. This simplifies assumptions removing the need to write comments about it, this also allows us to not return ArchiveBlockOutcome from constructor, maintaining previous API.

[teor2345]

Ok, this should do it the way you want now.

[nazar-pc]

Yes, thank you!