feat(blob): expose managed sharing in Python preparation API

[Xuanwo]

Prepared Blob writes can fail when Python's logical field carries empty Arrow extension metadata and a prepared field omits it. Accept those equivalent representations specifically for lance.blob.v2, including nested fields, while retaining strict checks for other extensions and nonempty metadata.

Expose BlobDescriptorArrayBuilder.append_managed in Python so callers can reuse an existing immutable payload or byte range within a table without assembling raw descriptor structs. Validation stays in the Rust core. The guide documents snapshot protection, base identity, and Managed-aware maintenance requirements.

Lifecycle regressions exercise two fragments sharing one payload, compaction, partial deletion, tagged-snapshot retention, and final reclamation for top-level and nested Blob fields in formats 2.2 and 2.3. The metadata regression was reproduced against the original implementation in both Python-to-Rust and Rust-to-Python directions.

Related to #7274. This does not add automatic deduplication, cross-table ownership, or media semantics.

[claude]

Claude Code Review

This repository is configured for manual code reviews. Comment @claude review for a one-time review, or @claude review always to subscribe this PR to a review on every future push.

_{Tip: disable this comment in your organization's Code Review settings.}

[lance-gatekeeper]

✅ Gate recommendation: approve.

The metadata regression and explicit-sharing lifecycle are supported by the focused Rust and Python tests, including compaction, tagged retention, and final reclamation. I found no blocker.