fix stale blob crossings in nightly soak

[feichai0017]

Summary

Fixes the nightly soak failures caused by stale blob crossings during DB/checkpoint churn and prepares the 0.7.2 patch release.

Root cause

• DB background auto-merge used the global GUID-only merge candidate queue. In a multi-tree DB, a queued parent can become unreachable from live roots while still sharing children with a live tree or snapshot, allowing a live child to be delete-fenced.
• The route cache allowed deeper parent anchors, but only validated the cached parent->child edge. It did not prove the parent was still reachable from the current root.
• When walkers saw a delete-fenced child through a stale crossing, the failure surfaced as pending-delete I/O errors or, under crash soak, acknowledged revision loss.

Changes

• Disable DB background auto-merge; DB compaction now stays rooted in live DB trees.
• Restrict route cache entries to root-child crossings and clear stale candidates on restart.
• Treat delete-fenced child pins as stale crossings and restart from the root instead of treating them as not found.
• Fence DB view capture through the same snapshot path as Tree view.
• Avoid merging snapshot-shared child blobs.
• Bump holt to 0.7.2, add changelog notes, and update crate repository/homepage metadata to NoKV-Lab/holt.

Validation

• cargo fmt --all
• cargo test --workspace --all-features --locked
• cargo clippy --workspace --all-features --all-targets --locked -- -D warnings
• cargo package --locked --allow-dirty
• cargo test -p holt --lib engine::route_cache::tests --locked -- --nocapture
• DB normal soak targeted run
• normal soak targeted run
• crash soak targeted run