fix: expose Iceberg work unit feed metrics

[alexanderbianchi]

Summary

Use the work unit feed as the shared metrics registry for Iceberg scans, whether execution is local or distributed.

• Give the local Iceberg feed a persistent metrics set.
• Register scan baseline metrics through IcebergDataSource::metrics(), which returns the feed’s shared set.
• Remove the separate data-source metrics field and codec-only initialization.

This exposes feed metrics without returning a fresh registry on each call, preserving later registrations such as DataFusion’s batches_split counter.

Validation

• cargo test -p datafusion-distributed-iceberg --features integration
• Formatting and diff checks passed.
• Temporary checks verified shared local registrations and local/distributed EXPLAIN ANALYZE metrics; these checks were removed and are not part of the PR.

[gabotechs]

This code here is just the codec. There's a chance that this code is not even reached in a normal execution, and this will still miss the coordinator-collected metrics.

Instead, we should be joining both DataSource and WorkUnitFeed metrics in the metrics() method:

impl DataSource for IcebergDataSource {
    ...
    fn metrics(&self) -> ExecutionPlanMetricsSet {
        let mut all_metrics = self.feed.metrics().clone_inner();
        all_metrics.extend(self.metrics.clone_inner());
        ExecutionPlanMetricsSet::from(all_metrics)
    }
    ...
}