feat: add ClickHouse read pool checkout failure telemetry

[amokan]

Overview

DBConnection sheds read pool checkouts under saturation (:queue_timeout) or when a queued request's deadline expires. Nothing counts them today, and the shed wait lands in read_pool.checkout.pool_time as if a connection had been obtained.

DBConnection already hands every failed checkout to the :log callback we pass on Ch.query/4, with connection_time nil. This matches that shape and emits a counter from the callback, where backend_id and the read cluster label are already in scope.

Key Changes

• New handle_read_pool_log/3 head matches a ConnectionError result with connection_time: nil and emits [:logflare, :clickhouse, :read_pool, :checkout_error] tagged backend_id, read_cluster, reason. A successful checkout always records a :prepare event first, so post-checkout errors never match
• The head skips the latency measurements and slow-checkout warning, so shed requests no longer inflate pool_time
• logflare.clickhouse.read_pool.checkout_error declared as a sum. reason mirrors DBConnection.ConnectionError.reason
• checkout_error and query_error descriptions state the overlap: a queue_timeout appears in both

Risks / Considerations

• Do not sum the two series checkout_error is per attempt and the pool-health signal; query_error is per query and the user-facing failure rate
• pool_time p99 drops under saturation Shed checkouts were the largest samples. Re-baseline any alerts on it