Kafka Connect: Add ConnectorMetrics instrumentation for observability

[kumarpritam863]

Summary

Adds a ConnectorMetrics facade that instruments the entire write and commit pipeline with 13 counters and gauges across three tiers: operational essentials, write path visibility, and schema/table management.

Today, the connector has zero metrics — operators monitor production deployments by parsing logs. This PR places instrumentation at every critical point in the data path so that when the Kafka dependency
moves to 4.0+ (KIP-KAFKA-15995 PluginMetrics API), all metrics automatically light up through the runtime's configured reporters (JMX, Prometheus, Datadog, etc.) with zero code changes.

On Kafka 3.9.x (current), counters are tracked in-memory via AtomicLong and accessible programmatically (e.g., for integration tests or custom health checks). The ConnectorMetrics.NOOP singleton ensures
zero overhead in tests.

Metrics catalog

Tier 1 — Operational Essentials

Metric	Type	Instrumentation Point
records-received-total	Counter	SinkWriter.save() — every batch received from put()
records-written-total	Counter	IcebergWriter.write() — each record successfully written
records-dropped-total	Counter	Reserved for routing drops (tombstones, empty route targets)
record-conversion-errors-total	Counter	IcebergWriter.write() catch block — type mismatches, parse failures
commit-total	Counter	Coordinator.commit() — every commit cycle initiated
commit-success-total	Counter	Coordinator.commit() — successful doCommit() completion
commit-failure-total	Counter	Coordinator.commit() — caught exception in commit cycle

Tier 2 — Write Path

Metric	Type	Instrumentation Point
data-files-written-total	Counter	IcebergWriter.flush() — from WriteResult.dataFiles()
delete-files-written-total	Counter	IcebergWriter.flush() — from WriteResult.deleteFiles()
active-writers	Gauge	SinkWriter.writers map size — registered via supplier

Tier 3 — Schema & Table Management

Metric	Type	Instrumentation Point
schema-evolutions-total	Counter	IcebergWriter.convertToRow() — schema update detected and applied
tables-auto-created-total	Counter	IcebergWriterFactory.autoCreateTable() — after successful creation
commit-timeout-total	Counter	Coordinator.process() — commit timed out waiting for workers

Design decisions

• ConnectorMetrics facade — All instrumentation calls go through one class (recordWritten(table), commitStarted(), etc.). This decouples metric recording from the metric backend. When PluginMetrics
  becomes available, only this class changes.
• enabled guard on all methods — The NOOP singleton has enabled=false, making every method a branch-predicted no-op. Enabled instances check once per call (branch on volatile boolean). No overhead
  when disabled.
• KIP-KAFKA-15995 ready — ConnectorMetrics.create() contains the exact try-catch pattern the KIP recommends (catch NoSuchMethodError | NoClassDefFoundError). When Kafka 4.0+ is available, metrics
  register under kafka.connect:type=plugins,connector=<name>,task=<id> automatically.
• No new connector config — The KIP handles all metric infrastructure (reporters, JMX registration, cleanup). No iceberg.metrics.* properties needed.
• tableName parameter on per-table methods — Even though table-scoped sensors aren't wired yet, the parameter is accepted now so call sites don't need to change when PluginMetrics adds per-table tag
  support.
• Duration tracking accepted but not stored — commitSucceeded(durationMs) and flushCompleted(tableName, durationMs) accept duration for future histogram/avg wiring.

[kumarpritam863]

@bryanck can you please check once.