[core] Support configured data directories throughout table IO

[JingsongLi]

Purpose

Support data-file.path-directory throughout Paimon table IO so PyPaimon can use Native planning, reading, writing, updating, and committing for tables with a configured data directory.

Previously, Rust built bucket paths directly under the table root. Using Native IO could therefore fail to find existing files or write files into a different layout from Java/Python.

Brief change log

• Resolve relative, normalized, absolute, and URI data directories using Java Path semantics, including Windows drives and UNC paths.
• Use the shared data root for append/PK/changelog/Blob writers, scans, managed Blob packs, update/delete paths, abort cleanup, dynamic-bucket hash indexes, and global-index source reads.
• Preserve the explicit relative bucket path for external deletion vectors instead of assuming every bucket lives under the table root.
• Make physical/referenced file statistics aware of relocated data and sidecar files, including directories outside the table root.
• Reject an empty directory and preserve the stored layout during table copies and catalog option changes. Catalog alterations use the existing index-layout immutability rule, including before the first snapshot, to avoid stranding an already-created writer.
• Add end-to-end coverage for relocated files, updates/upserts, repeated deletes, history, cleanup, Blob payloads, dynamic buckets, and sidecar statistics.

Tests

• 35 integration tests passed: data_file_directory_test, table_update_paths_test, and table_update_test.
• 76 focused unit tests passed: partition/path utilities, directory immutability, external paths, referenced/physical statistics, dynamic bucket assignment, and index paths.
• cargo clippy --locked --all-targets --workspace --features fulltext,vortex -- -D warnings passed.
• cargo fmt --all --check passed.
• Built the Python binding locally and ran the companion PyPaimon Native regression selection with all five Native CI flags: 1368 passed, 2 skipped, and 62 subtests passed (PyArrow 18.1).

API and Format

Adds a CoreOptions accessor and options-aware file-statistics entry points. Existing statistics entry points retain their default-directory behavior. No file-format or manifest-format changes; schema, snapshot, manifest, and table-level index metadata remain rooted at the table location.

Documentation

Includes API documentation and Java-derived path expectations in regression tests. The option already exists in Java/PyPaimon.

[leaves12138]

Reviewed commit 2a46e49fe7433165bf066c1e4c6abac8cf3cec60 together with apache/paimon#10249.

Fixed and pushed the findings from this review:

• Preserve the ./ prefix when normalization exposes a colon in the first relative path component. Windows must not reinterpret that component as a drive root; this fixes the previous Windows unit-test failure.
• Preserve an empty path when a child such as . or x/.. resolves against a scheme/authority-only parent, matching Java Path instead of introducing an extra slash.
• Added common and explicit Windows-mode regression cases, including Windows-mode coverage runnable on Linux.

Validation of this revision:

• cargo test --locked -p paimon --lib: 3,410 passed, 6 ignored.
• data_file_directory_test, table_update_paths_test, and table_update_test: 35 passed.
• cargo fmt --all -- --check: passed.
• Built the Python native extension from this exact Rust revision and tested it with the paired Python changes: 1,748 passed, 4 skipped, and 143 subtests passed with PyArrow 18.1.0 and native planning/reading/writing/updating/committing enabled.

No remaining blocking code-review findings for the paired revisions. The new cross-platform CI run is still in progress; this approval does not replace the required CI checks. LGTM.