Implement ReminderSchedulerService orchestration logic

Use a Stream> from the repository to enable lazy batch loading, or use offset-based pagination with a while loop and a hasMore flag — the latter is simpler and easier to test. Model EvaluationResult as a sealed class with three subtypes (Remind, Escalate, NoAction) so the routing switch is exhaustive and the compiler enforces all cases. Accumulate run results using a mutable _RunAccumulator helper class (private to the implementation file) that exposes increment methods — this keeps the main method readable. Wrap each per-assignment dispatch in a try/catch that catches Object and logs the error with the assignment ID before continuing.

Do not use Future.wait() for the entire assignment list at once — only within each batch — to bound concurrency and memory.

Unit tests using flutter_test with all dependencies mocked via mocktail. Required test scenarios: (1) empty assignment list returns SchedulerRunResult.empty(); (2) all assignments return EvaluationResult.remind → remindersDispatched count equals assignment count; (3) all assignments return EvaluationResult.escalate → escalationsDispatched count correct; (4) mixed results accumulate correctly across remind/escalate/noAction buckets; (5) one dispatch failure does not abort remaining assignments and failed assignment does not increment success counters; (6) batch size respected — verify repository paginated calls match expected batch boundaries. Integration test (optional, against Supabase local emulator): full run with 20 seeded open assignments across multiple batches.