Spark V2 scans report no input bytes to Spark's metrics #99

Spark V2 Vortex scans leave a stage's inputBytes / bytesRead at 0, so the Spark UI, event logs and cost tooling can't see how much a scan read.

The likely cause: reads run on Vortex's native worker threads through PooledReadable. Spark fills a task's input metrics from Hadoop FileSystem statistics on the task's own thread, so it never sees reads made on other threads.

The connector's tests measure bytes read through a test-only hook (VortexNestedSchemaPruningIoTest.BYTES_READ); nothing reports them to Spark at run time.

Proposal.

• Count bytes read per partition reader inside PooledReadable (or VortexIo).
• Report them through PartitionReader.currentMetricsValues() as custom metrics, and also add them to the task's inputMetrics when a TaskContext is present, so bytesRead reflects the scan.