fix: make the native Iceberg per-I/O timeout configurable

[comphead]

Which issue does this PR close?

Closes #6124.

Rationale for this change

Native Iceberg scans fail on OpenDAL's hard-wired 10s per-I/O timeout (io operation timeout reached), and nothing in Comet could change it. apache/iceberg-rust#3263 made the timeout configurable through the opendal.io-timeout-ms FileIO property.

What changes are included in this PR?

• Bump iceberg and iceberg-storage-opendal from bb1e4a4 to af1da4c, the merge commit of feat(storage-opendal)!: make the per-IO-operation timeout configurable iceberg-rust#3263 (22 commits).
• Add spark.comet.iceberg.ioTimeout (default 30s). The native Iceberg scan and write pass it to iceberg-rust as opendal.io-timeout-ms.
• Add opendal. to the native STORAGE_PROPERTY_PREFIXES so the property reaches FileIO. The merged upstream key is opendal.io-timeout-ms, not the client.io-timeout-ms the issue expected, so this is needed.
• API updates the bump requires: FileWrite::close now returns FileMetadata, and UnboundPartitionField fields are private.

How are these changes tested?

• New native unit test io_timeout_reaches_the_file_io: the key matches iceberg-rust's OPENDAL_IO_TIMEOUT_MS and passes the storage-prefix filter.
• Every native Iceberg scan and write now carries the property, so the existing Iceberg suites run with it set.

[sunchao]

Summary

• Prior state and problem: Native Iceberg operations used OpenDAL’s fixed 10-second per-I/O timeout, which Comet could not configure.
• Design approach: Upgrade iceberg-rust and pass spark.comet.iceberg.ioTimeout through the existing scan/write property channel.
• Correctness / compatibility analysis: No introduced P1/P2 issues found within this review. Reviewed all nine base-relative files and relevant upstream reader, storage, partition and writer changes. Compared duration parsing against Spark 3.4–4.2 sources and relevant behavior against supported Iceberg Java versions. Existing reviews, comments and threads were empty.
• Key design decisions: The positive-duration setting defaults to 30s. FileIO cache keys include the setting. The implementation reuses existing configuration and serialization infrastructure without adding per-row work.
• Implementation sketch: Forward opendal.io-timeout-ms, admit the opendal. property prefix, preserve returned FileMetadata, and migrate partition-field construction to the upstream builder.
• Behavioral changes worth calling out: Compared with branch-1.1 at 992c806a7e38c2e88bd018aa5774164b0850e1fa, the per-I/O default intentionally increases from 10 to 30 seconds and becomes configurable. OpenDAL’s separate 60-second control-operation timeout remains unchanged.
• Suggested improvements: No additional changes meeting the P1/P2 reporting threshold.

Reviewed full SHA 22abdf4151c6451bdbfa663e5be1f3166954e5c7 against 33f21da181c14df7fa3e6a3104c3d7215810147e. Routed skills: review-comet-pr, review-comet-expression-pr, and review-comet-iceberg-write-pr.

Validation: All 121 native Iceberg tests and the changed planner test passed. A disposable probe confirmed an actual OpenDAL read timeout at 25 ms and successful FileIO round trips with close metadata. The default-feature build encountered missing jni.h, so native tests ran with HDFS disabled.

Exact-head CI: 17 checks succeeded, two remained running, and 14 were skipped. Native build and Rust CI were pending. Spark SQL and Iceberg integration suites were skipped. Local JVM integration suites and real object-store validation were not run, so merge-readiness validation remains outstanding.