Add basic remote storage test

[SomeoneToIgnore]

Part of #755

All fixes and improvements are put as different commits in the same PR for simplicity and better context, but I can split them off, if needed, but the changes are rather small combined IMO and depend on each other anyway, so I'd left it all as is now.

Fixes:

• sync loop blocking and not letting pageserver to shutdown properly

Improves:

• Zenith CLI now passes RUST_LOG env var to safekpeer and pageserver CLI commands, python tests can also specify the same env var to automatically propagate it into the CLI or programmatically set a rust_log_override property to enable the same for a certain test
• Zenith CLI now accepts --pageserver-config-override params that get passed into pageserver as -c (--config-override), which is used to specify remote storage params to pageserver via Zenith CLI.
  Also, ZENITH_PAGESERVER_OVERRIDES env var can be used to set those config values globally for all python tests (so that all existing tests could be run with overrides, e.g. remote storage enabled)

Adds:

• a bit more tests on new pageserver's timeline http endpoints
• a way to programmaticaly enable remote storage and RUST_LOG overrides in python tests
• simple remote storage timeline restoration test

The remote storage test is not finished yet, it requires more fixes to properly read the WAL after the timeline data got restored, I will do that later: right now the test is useful already, since it brings the remote storage code needed by #995 and tests that pageserver is capable to upload, download and load a timeline into its memory correctly.
Also, all remote storage tests now use local fs as a storage, so future improvements are needed to bring some form of S3 into the tests.

[hlinnaka]

The config option changes look good to me, I didn't look at the rest of the patch

[hlinnaka]

What does RUST_LOG_KEY do?

Would it make sense to pass through all env variables with RUST_* prefix?

[SomeoneToIgnore]

What does RUST_LOG_KEY do?

That's just a variable to hold the RUST_LOG value to use it twice on the next lines. const forces it to BE_NAMED_THIS_WAY , so might be a bit confusing. I can define it as let rust_log_key = ..., would that reduce the confusion?

Would it make sense to pass through all env variables with RUST_* prefix?

I'm not aware of any others that start with such prefix, actually.
Seems that the RUST_LOG one emerged from env_logger crate and still isn't a standard really, since has to be supported in every logging backend separately. For instance, the slog one we've used before tracing did not support such variable.

Lastly, the only documentation about any rust-related env vars I was able to find is about cargo:
https://doc.rust-lang.org/cargo/reference/environment-variables.html

So, I'd rather keep the RUST_LOG one only for now.

[LizardWizzard]

Nice! Looks good to me, left few minor comments

I plan to use pageserver related utilities in a tenant migration tests