feat(buffer): implement Buffer with pluggable storage

[prosdev]

What This PR Does

Implements Issue #5 (Buffer) with per-partition buffering, size-based and time-based flushing, and pluggable storage backends.

Key Features

• Pluggable Storage: BufferStorage protocol enables in-memory (default), disk, or Redis backends
• Per-Partition Isolation: Each partition has its own buffer for isolation and ordering
• Size-Based Flush: Automatically flushes when buffer reaches configured size (default: 100 events)
• Time-Based Flush: Background task flushes all partitions every timeout seconds (default: 5s)
• Graceful Shutdown: stop_flusher() ensures no events are lost on shutdown
• Defensive Max Size: Prevents OOM with max_size limit (default: 1000 events/partition)
• Retry-Flush Strategy: Attempts one flush before raising BufferFullError (production CDP pattern)

Commits

1. feat(buffer): add Buffer with pluggable storage and retry-flush

   • Create BufferStorage protocol and InMemoryBufferStorage
   • Implement Buffer class with enqueue() and size-based flush
   • Add retry-flush-before-failing strategy
   • Add BufferFullError for defensive max_size limit
   • Comprehensive tests (97% coverage on buffer.py, 95% on buffer_storage.py)

2. feat(buffer): add time-based flush and graceful shutdown

   • Implement start_flusher() for background task
   • Implement _run_flusher() for periodic flushing
   • Implement flush_all() to flush all partitions
   • Implement stop_flusher() with graceful shutdown (no data loss)
   • Add tests for time-based flush, shutdown, and concurrency
   • All 11 tests passing (96% coverage)

3. docs: mark Task 8 complete in tasks.md

   • Update tasks.md to reflect completion

Test Coverage

• 11 unit tests passing
• 96% coverage on buffer.py
• 100% coverage on buffer_storage.py
• Tests cover: initialization, enqueue, size-based flush, time-based flush, graceful shutdown, concurrency, error handling, per-partition isolation

Design Decisions

See notes/projects/eventkit-impl/005-buffer-batching.md for detailed analysis:

• Why per-partition buffers (vs. single buffer)
• Why asyncio.Lock is necessary
• Why background task for time-based flush
• Pluggable storage design
• Max buffer size strategy

Closes #5