Directory structure impl

[coolreader18]

Description of Changes

Implements the new directory structure proposal, using strongly-typed newtypes to represent paths. All places in the database/server/cli that previously would have been passing around PathBufs and .join()ing them now instead pass around these types, calling methods on them that internally call .join() and return another strongly-typed child node in the directory hierarchy.

Changes from the proposal implemented here:

• the replicas directory in the data-dir is flattened from having database files be at replicas/$database_identity/$replica_id/database/..., to just replicas/$replica_id/...
• all environment variables are removed, besides a few (SPACETIMEDB_DISABLE_DISK_LOGGING, SPACETIMEDB_TRACY, SPACETIMEDB_FLAMEGRAPH{,_PATH}) that seem to be just for debugging purposes. I can turn them into command-line args if that's desired.
• I split the ~/.spacetime directory across the xdg base directories on unix-likes (so ~/.config/spacetime, ~/.local/share/spacetime, etc). On windows it's moved to %LocalAppData%/SpacetimeDB
• Split the config.toml into cli.toml and data-dir config.toml, and implemented data-dir/metadata.toml.
• data-dir/wasmtime_cache -> data-dir/cache/wasmtime
• module logs are dated instead of just 0.log (though a proper logrolling mechanism hasn't been implemented - that likely needs further work), database logs are in data-dir/logs instead of /var/logs/spacetime.
• data-dir/spacetime.pid lockfile, so multiple processes can't use the same data-dir at the same time.
• All the root paths (cli-{bin-file,bin-dir,config-dir}, data-dir) are now bundled in the SpacetimePaths type. server processes like standalone only interact with the data-dir.
• the CLI has 2 new flags for the base command itself, not on subcommands - spacetime --root-dir and spacetime --config-path, to set a custom root directory for the spacetime installation and to specify a custom cli.toml, respectively. this was not in the proposal itself but something like this had to be implemented for our tests to still work.

Changes that are not directly related to that, but facilitate it:

• the spacetimedb::auth::JwtKeys type, which deduplicates jwt-key-file code across our server implementations. This is something I have less good motivation for, and I could try and revert if desired.
• Host::try_init now takes &HostController, instead of like 5 different args that were just fields on HostController passed individually. I was adding a 6th arg and clippy was getting mad at me for having too many parameters.
• the wasmtime Engine and Linker are now stored in a WasmtimeRuntime struct. Since we have to give Engine the {data-dir}/cache/wasmtime path, it made more sense for that to just get eagerly initialized by HostController, who has access to the data-dir path, instead of threading data_dir all the way through to wasmtime::make_actor.
• relational_db::test_utils::TempReplicaDir, which merges a TempDir and spacetimedb_paths::server::ReplicaDir.
• Added spacetimedb::startup::TracingOptions got added - this is so we wouldn't read environment variables in core anymore.

[mamcx]

Some other nits:

• It will be nice to see a test like in my pr
• Also, to add in the lib a copy of the proposal some the reviewers and future readers see the big picture.
• In general, no code outside the Paths should build the layout, names, etc.

[gefjon]

Many thanks for the in-depth PR description. That makes review much easier.

[jsdt]

This seems like an improvement to me, though it would be a good time to add some more comments. I also wrote a few unit tests that you could add: noa/directory-structure...jsdt/lock-test

[bfops]

CLI changes LGTM