fix testing packages with multiple targets

[duncanawoods]

This PR fixes #18955 and other bugs related to RA executing the wrong test and/or assigning results incorrectly.

It makes test execution aware of the difference between packages and targets.

It removes the need to use hack_recover_crate_name to match test output to tests.

• request::handle_run_test
  • no longer tries (and fails) to reduce a list of tests to a single pattern
• test_runner::CargoTestHandle
  • no longer runs patterns against the workspace and now always runs tests in the correct target
• CargoActor
  • is extended to take a parser that has the required context for parsing tool output correctly
• main_loop
  • every CargoTestMessage in the event queue now contains the required information to assign to correct test

[HKalbasi]

Ah sorry I forgot this. Now that the hack_recover_crate_name is gone, this actually reduces the amount of workarounds so I'm happy to merge this if you remove the merge commits.

[duncanawoods]

Oh that's great, thank you @HKalbasi !

Do you have any thoughts about adding some testing to this feature?

Having seen the dizzying number of permutations of how tests are defined and how they might be invoked I'm nervous about regressions. I'm thinking of an integration test using a sample workspace that tries to define every type of test and we might be able to simulate vscode interaction with "code --command" .

[HKalbasi]

If you want to add some end to end tests, look at https://github.com/rust-lang/rust-analyzer/blob/master/crates/rust-analyzer/tests/slow-tests/main.rs which creates a cargo project and communicate with rust-analyzer using lsp. But those tests slowdown CI so I would suggest having at most one test about test-explorer there (which can test many operations, most of the time is spent on the loading) and test other corner cases using unit tests that takes some hard coded json of cargo and/or assert text of the cargo commands instead of executing them.