[V2] Prefix compression for keys

[bravo1goingdark]

Summary

Implement prefix compression to reduce memory usage for keys that share long prefixes (e.g., user:12345:email, user:12345:name).

Motivation

For sorted workloads with keys sharing common prefixes (common in LSM-tree use cases like user:{id}:{field}), storing only the suffix can significantly reduce memory usage.

Implementation Notes (from design doc)

Store only the suffix after a shared prefix:

struct NodeHeaderCompressed {
    shared_prefix_len: u32,  // bytes shared with predecessor's key
    key_suffix_len: u32,
    // ...
}

This requires the predecessor to be stable (it is, since nodes are never moved). Implementation adds complexity to find_less and comparison.

From design doc

See LOCKFREE_SKIPLIST_DESIGN.md lines 391-403

Priority: medium