Expose Inserter metrics / debug info for performance monitoring

### Use case
Now, `clickhouse.Inserter` like `black box` - it's hard to see what's happening with performance

I want to be able to:
- monitor how many rows `Inserter` actually has buffered at any moment.
- *when* and *why* a flush is triggered (manual `commit()`, `with_period()`, `with_max_rows()`, etc.).
- see the size of each batch of data actually sent to ClickHouse.
- measure the local-to-server round-trip time per flush (commit RTT).
- track total rows sent and batches committed over time.

### Describe the solution you'd like

expose client-side flush/debug metrics in prometheus format like:

- buffered_rows / buffered_bytes
- total_rows_sent / total_batches
- last_batch_rows / last_batch_bytes
- last_batch_rtt_seconds
- last_flush_reason

### Describe the alternatives you've considered

add inserter function:
- buffered_rows() → current buffered row count
- buffered_bytes() → estimated serialized size in bytes
- on_flush(callback: Fn(&FlushInfo)) → hook fired before/after every flush
- with_debug_logging(enabled: bool) → enable structured debug logs for Inserter

### Additional context

Start by locating clickhouse.Inserter and reviewing how buffering and flushes currently work. Clarify the metrics API, flush-reason semantics, and Prometheus format before implementation; done should include agreed metrics for buffering, batches, rows, RTT, and flush reasons.

Written by the indexing model from the issue text.