feat(eval): add trial count support

[Abhijeet Prasad (AbhiPrasad)]

Run eval cases multiple times with a global TrialCount option and allow each case to override the default. Expose the zero-based TrialIndex through task hooks so tasks can distinguish repeated executions.

Public API additions:

// eval.Opts
TrialCount int

// eval.Case
TrialCount int

// eval.TaskHooks
TrialIndex int

Example usage:

result, err := evaluator.Run(ctx, eval.Opts[string, string]{
    Experiment: "trial-count-demo",
    Dataset: eval.NewDataset([]eval.Case[string, string]{
        {Input: "hello", Expected: "HELLO"},
        {Input: "world", Expected: "WORLD", TrialCount: 3}, // per-case override
    }),
    Task: eval.T(task),
    Scorers: []eval.Scorer[string, string]{scorer},
    TrialCount: 2, // default trials per case
})

task := func(ctx context.Context, input string, hooks *eval.TaskHooks) (eval.TaskOutput[string], error) {
    log.Printf("running trial %d", hooks.TrialIndex)
    return eval.TaskOutput[string]{Value: strings.ToUpper(input)}, nil
}