[squiidz]

What

aws_dynamodb_cdc retries a closed shard's failed completion write on every shard refresh, so the shard's tracker entry no longer lingers against max_tracked_shards until the shard ages out of the stream.

Why

finalizeShard left finalizing=true when MarkShardCompleted failed, so the completion was never retried and the entry stayed tracked for up to 24h (found while working on CON-599, #5001).

How

• finalizeShard clears the finalizing claim when the completion write fails, so a later caller can take it again.
• New RecordBatcher.RetryFinalize claims and finalizes every closed shard with no inflight batches.
• refreshShards and refreshTableShards call it once per refresh cycle via retryShardFinalization, right after forgetShards; failures are logged at Warn and retried next cycle.
• Stacked on CON-599: aws_dynamodb_cdc: stop re-reading exhausted closed shards #5001 (jira/CON-599): the finalization code this fixes only exists there.

Testing

• Unit tests: 5 regression tests fail without the claim release and pass with it
• go test -race ./internal/impl/aws/dynamodb/... green
• golangci-lint 0 issues; task fmt clean; task build:redpanda-connect succeeded
• Integration: dynamodb package with -tags integration against Docker, passed in 145s
• Skipped: dynamodb-cdc end-to-end scenario (enterprise connector, no license file)

Risks

Behavior change only after a failed completion write: one extra DynamoDB write per affected shard per refresh cycle plus a Warn log each cycle while the store keeps failing. No config or spec changes.

Jira: https://redpandadata.atlassian.net/browse/CON-601