Research: specify checkpoint storage and durability

[joaomj]

Parent map: #1

Decision to resolve

How should exact-undo checkpoint blobs and manifests be stored, synchronized, bounded, retained, encrypted, and garbage-collected?

Inputs

• Research: define durable turn recovery semantics #3 durable turn recovery semantics
• Research: define worktree, undo, and completion semantics #6 worktree and exact undo semantics

Output

A storage and durability contract covering atomic manifest/blob publication, coherent workspace scans, capacity failure, access control, retention, and recovery verification.

Resolves when

An implementation can persist and recover a complete checkpoint without silently skipping data or overstating durability.