Improve engine_correctness_tests for maintainability

[V0ldek]

Is your feature request related to a problem? Please describe.

1. Cases are confusingly named, describing files and queries, but not what they're meant to test. When a test fails it's impossible to tell what exact part of the system failed, which is important.
2. The JSON files themselves also have incosistent names, sometimes describing the document, other times the query use-case they want to stress.
3. There is no consitency in which files are tested for prettified vs compressed versions.
4. Some tests verify only counts, some verify both counts and byte indices, resulting in inconsistencies and duplication of test cases.

Describe the solution you'd like

1. All cases should describe what they test in high-level, declarative means. Each test cases has a reason for its existence.
2. JSON file names should describe their structure, not what cases they're used for. A slight exception for this is the wikidata dataset, for which "structure" simply means "it's the wikidata real-life dataset".
3. There should be automatic verification of that, and all tests should have prettified and compressed versions. A script should automatically create the missing counterpart, integrated with the justfile. Part of just verify should check that each test has a counterpart.
4. We should always verify byte indices, since correct indices imply correct counts. In case of large result sets we should have snapshot files with the output.

[github-actions]

Tagging @V0ldek for notifications