Flink: Fix test assumption which can produce flakiness

[mxm]

Under high load due to limited CI resources and parallel test execution, the test pipeline can run multiple Flink checkpoints before all records have been processed. This produces a slightly different but not semantically wrong outcome.

In the below reported test failure, there won't be a positional delete because the Flink checkpoint cuts off the writer optimization to convert an equality delete to a positional delete.

TestDynamicIcebergSink > testCommitsOncePerTableBranchAndCheckpoint() FAILED
    java.lang.AssertionError: 
    Expecting map:
      {"added-data-files"="1", "added-delete-files"="1", "added-equality-delete-files"="1", "added-equality-deletes"="1", "added-files-size"="1128", "added-records"="1", "changed-partition-count"="1", "engine-name"="flink", "engine-version"="2.1.0", "flink.job-id"="7a390dfdf54bc3d7028708e2aa46c54c", "flink.max-committed-checkpoint-id"="2", "flink.operator-id"="27651a86dc01e60882f6b181172fad4c", "iceberg-version"="Apache Iceberg aa63060 (commit aa630608f79d06c1c3929801fd7266b082170758)", "total-data-files"="5", "total-delete-files"="2", "total-equality-deletes"="2", "total-files-size"="4478", "total-position-deletes"="0", "total-records"="6"}
    to contain entries:
      ["total-equality-deletes"="1",
        "total-position-deletes"="1",
        "total-records"="6"]
    but the following map entries had different values:
      ["total-equality-deletes"="2" (expected: "1"), "total-position-deletes"="0" (expected: "1")]
        at org.apache.iceberg.flink.sink.dynamic.TestDynamicIcebergSink.testCommitsOncePerTableBranchAndCheckpoint(TestDynamicIcebergSink.java:1005)

The important part of the test is to ensure comitting once per checkpoint which is asserted further below.

This closes #15139.

[singhpk234]

Looks reasonable to me !

[kevinjqliu]

ty!