Add AsyncLoader with dependency tracking and runtime prioritization

[serxa]

This tool is needed to make async table loading possible in #43424.
Please, read a comment about AsyncLoader class in the source code for implementation details.

Here is how it is going to be used in the follow-up PRs. But I want to merge AsyncLoader first. The idea is to load the dependency graph created by TablesLoader into AsyncLoader as a graph of LoadJobs. There will be at least 4 types of LoadJobs:

1. Load table (including not outdated parts).
2. Startup table (depend on 1, has initial low priority to allow loading of all table before startups to delay the start of merges/mutations).
3. Startup DDLWorker for ReplicatedDatabase (depends on 2 for all tables, also has low priority because depends on low-priority jobs).
4. Startup Server job (no-op job) clickhouse-server will wait on this job before starting servers (depends on 2 for all tables that should be loaded before server start.

On incoming queries, we will wait for every involved table to be started (job type 2). Important: to speed up the loading of these tables, we will prioritize all corresponding jobs (with their dependencies) using AsyncLoader::prioritize(job, new_priority).

cc @alesapin @tavplubix

Changelog category (leave one):

• Not for changelog (changelog entry is not required)

[serxa]

Stateless tests (asan) [2/4] — fail: 1
00814_replicated_minimalistic_part_header_zookeeper - #43983

[serxa]

Unit tests (msan) — Exit on signal. PASSED record not found.

msan found a problem with JobFailure unittest: https://pastila.nl/?0060f2ab/915507a84a7d00c58c8f4f12bf83efc6

Reproduced. Again, I'm not sure what is the reason. But now we can see there is access to uninitialized variable

UPD. should be fixed by #49316

[serxa]

@rschu1ze Could you please take a look at this https://s3.amazonaws.com/clickhouse-test-reports/48923/1263c58f648dd8c3b111f82b2e50123c40789719/unit_tests__msan_/run.log

I thought it should be fixed by #49316. But it is not, for some reason. Or the "Update branch" button is broken.

[rschu1ze]

@serxa let me try to reproduce locally ...

[rschu1ze]

Reproduces easily, checking ...

[rschu1ze]

@serxa Unlike first suspected, the failure is definitely unrelated to #49316. And I am pretty sure it is a false alarm. I tried to suppress it yesterday with __msan_unpoison (in demangle.cpp) and -fsanitize-blacklist but without success yet. I will need to check phdr_cache.cpp closer.

To un-block this PR, feel free to disable the failing test for now(TEST(AsyncLoader, DISABLEDJobFailure)), I'll follow up in a separate PR.