fix: durably reuse vacuumed data pages

[pathscale]

Fixes #31.

Persist whole data-page free ranges behind an ordered persistence barrier. Vacuum queues each reclamation marker before exposing that page to in-memory allocators, and later writes consume durable ranges before writing row bytes. This stops repeated .wt.data growth across reloads without claiming online file truncation.

Regression coverage:

• test_vacuum_on_persisted_table_survives_reload proves free space survives reload, the next insert does not grow the file, and a second reload/reuse cannot overwrite the first reused row.
• vacuum_reclamation_waits_for_preceding_row_moves proves FIFO ordering across row batches and the reclamation barrier.

Stacked on #51, which is stacked on #48.

Validation:

• cargo fmt --check
• cargo clippy --all-targets -- -D warnings
• cargo test (151 library, 356 integration with 2 ignored, 2 doctests)