Separate profiler out of the user interface

[krassowski]

Addresses #45

This allows downstream to run benchmarks programatically, including via playwright if jupyterlab app is exposed.

Code changes:

• the main public API entry point, IUIProfiler was extended to offer the following methods and properties:

  interface IUIProfiler {
    addScenario(scenario: IScenario): void;
    runBenchmark<T extends IOutcome = IOutcome>(
      scenario: {
        id: string;
        options: JSONObject;
      },
      benchmark: {
        id: string;
        options: JSONObject;
      }
    ): Promise<IBenchmarkResult<T>>;
    readonly scenarioAdded: ISignal<IUIProfiler, IScenario>;
    abortBenchmark(): void;
    readonly progress: ISignal<IUIProfiler, IProgress>;
    readonly benchmarks: IBenchmark[];
    readonly scenarios: IScenario[];
  }

• new plugin @jupyterlab/ui-profiler:user-interface was added which relies on IUIProfiler provided by @jupyterlab/ui-profiler:plugin
• all interfaces required for core profiler were moved to tokens.ts and are now exported as public API members
• the user interface react widget was renamed from UIProfiler to UIProfilerWidget
• the IUIProfiler is now implemented in new UIProfiler class
• IBenchmarkResultBase, IBenchmarkData, and type BenchmarkResult were merged into a single simplified IBenchmarkResult
  • IOutcome is now stored in IBenchmarkResult under .outcome field rather than (confusingly) .result as before; this is a soft deprecation, result loader supports the old format for now.

[github-actions]

👈 Launch a Binder on branch krassowski/ui-profiler/separate-user-interface

[krassowski]

There is a trade-off in runBenchmark() definition which currently does not have strict types (but does not require retrieval of specific scenarios/benchmarks directly); the alternative would for the downstream user to provide the scenario and benchmark objects with pre-set options.