perf: testing iceberg-rust ArrowReader optimizations [iceberg]

[mbutrovich]

Which issue does this PR close?

Closes #.

Rationale for this change

What changes are included in this PR?

• Change iceberg-rust dependency to experimental branch at https://github.com/mbutrovich/iceberg-rust/tree/reader_perf that includes:
  • Parquet metadata caching
  • Serialize file size from Iceberg DataFile to native FileScanTask to avoid a stat() call
  • OpenDAL operator caching
  • Parquet metadata prefetch (512KB)
• Add a config to set iceberg-rust's data_file_concurrency_limit, defaults to 1 since tests without an ORDER BY fail when you increase the value.

How are these changes tested?

Existing tests. I also benchmarked and saw the flame graphs stacks associated with stat() disappear, and other stacks shrink.