Summary

• enable writing table metadata for format v3 and raise the supported format ceiling to v3
• add ManifestWriterV3 and ManifestListWriterV3, including v3 manifest-list first_row_id assignment
• wire snapshot commits to populate v3 first_row_id and added_rows from manifest-list row-id assignment
• allow transaction format upgrades to v3 and initialize next_row_id=0 for new/upgraded v3 tables
• update unit and integration tests to assert v3 write success (including nanosecond schema and row-lineage assertions)

Implementation notes

• DEFAULT_READ_VERSION in manifest handling is now 3 so v3 fields (including ManifestFile.first_row_id) are preserved when reading/re-writing manifests.
• v3 manifest-list row-id assignment follows Java behavior: assign only for data manifests missing first_row_id, and advance by existing_rows_count + added_rows_count.

Follow-up scope (tracked by #1551)

• This PR intentionally targets the v3 core writer path for append/delete/overwrite commits.
• Remaining v3-related scope appears broader, including equality-delete execution support, deletion-vector execution semantics, and wider catalog/runtime parity coverage (for example, Hive paths).
• If maintainers prefer, I can add part of this missing scope to this PR; otherwise I can open a focused follow-up PR linked to #1551.

Validation

• ./.venv/bin/python -m pytest -q tests/table/test_metadata.py tests/table/test_init.py tests/utils/test_manifest.py
• uv run python -m pytest -q tests/integration/test_reads.py -k upgrade_table_version
• uv run python -m pytest -q tests/integration/test_writes/test_writes.py -k "test_nanosecond_support_on_catalog or test_v3_write_and_read_row_lineage"

Related: #1551