Add async single-flight coalescing (dogpile prevention)

[toloco]

Summary

• When multiple async coroutines miss the cache for the same key concurrently, only one computes the result — the rest await an asyncio.Event and reuse it
• Works for both memory and shared backends, default on for all async cached functions
• Handles error recovery (waiters retry via leader re-election loop) and cancellation cleanly

Closes #16 (single-flight portion; non-blocking waits deferred as low-priority).

Test plan

• test_async_single_flight — 10 concurrent coroutines, assert exactly 1 computation
• test_async_single_flight_different_keys — no false coalescing across keys
• test_async_single_flight_error_recovery — leader fails, waiters recover
• test_async_single_flight_cancellation — leader cancelled, waiters recover
• test_async_single_flight_shared — works with shared backend
• test_async_concurrent assertion tightened from >= 3 to == 3
• Full test suite (77 tests) passes
• Benchmarks show no regression

🤖 Generated with Claude Code