Capture and inspect all logs in Catch2 tests, fix race

[fknorr]

This PR completely redefines how logging is handled inside unit tests.

• fixes a long-standing data race when capturing and inspecting logs inside a test
• silences logs by default, printing them post-mortem when a test fails
• detects if an unexpected warning or error is logged and fails the test

It works by unconditionally capturing logs in the global thread-safe test_utils::log_test_capture. This allows the test code to query captured messages, and will echo the captured logs through a global Catch2 hook if the test fails. By default, this mechanism will fail any test that logs a warning or error, but it provides functions to relax that behavior either explicitly through test_utils::allow_max_log_level or implicitly through runtime_fixture / device_queue_fixture which permit system-dependent warnings to appear.

The new mechanism is fully independent of the CELERITY_LOG_LEVEL environment, both in which messages it captures and also which ones it dumps in case of a test failure.

Some refactorings and fixes happened alogn the way:

• Simplified log.h: log_context and its printing was unnecessarily complex
• Moved std::tuple utils from log.h to utils.h
• Make logging in runtime and from utils::panic independent of LOG_LEVEL when called from tests
• Fixed some uninit-read bugs in tests that were surfaced by the new log-level restrictions

[github-actions[bot]]

Check-perf-impact results: (4c65f1399a47e0eb1340f63004745b17)

:question: No new benchmark data submitted. :question:
Please re-run the microbenchmarks and include the results if your commit could potentially affect performance.

[fknorr]

I agree, finer control over which warnings to allow would be great in a follow-up PR!

I have created an issue about this.