storage: harden concurrent checkpoint recovery

[aikins01]

Hardens concurrent writes and checkpoint recovery against corruption, races, and unsafe continued operation after interrupted checkpoints.

• preserves out-of-place node, relationship, CSR, and free-space checkpoint state across failure and recovery
• makes active/frozen WAL and shadow cleanup ordering explicit, including cleared or missing shadow recovery
• enforces first-writer-wins catalog updates and synchronizes catalog version-chain publication
• adds native and TSAN regressions for concurrent DDL, checkpoint failure, WAL recovery, CSR preservation, and agent-memory workloads

This remains a draft while directory-entry durability, commit/WAL failure semantics, checkpoint resource leaks, crash-injection coverage, and sticky rotated-checkpoint behavior are resolved or explicitly classified.