Summary

Moves compressed on-disk RDB (the RDB_ENC_COMPRESSED encode/decode path, R2.6.1–R2.6.4) from v1 to v2. Docs-only.

Rationale

Compressed on-disk RDB is a local-snapshot optimization — smaller/faster RDB file and faster warm-up back to the compressed state after a restart. It is not a correctness requirement and not a sync-speed improvement: full-sync replication is uncompressed in v1 regardless (R2.6.8).

v1 is already correct and memory-safe without it:

• Every RDB save target (disk, full-sync, AOF preamble, DUMP/MIGRATE) emits uncompressed, routing each value through objectGetUncompressedView at save time (PR . valkey-io/valkey#24). Verified: rdbSaveStringObject (rdb.c) and rioWriteBulkObject (aof.c:1909) both decompress.
• Decompression is transient and per-value — one throwaway scratch sds at a time; the live compressed robj is never mutated. A full-keyspace save never decompresses everything at once; peak overhead is a single value ≤ compression-max-value-size. (This also answers the memory-overflow concern: no overflow risk.)
• On load there's nothing compressed to read; the sweeper re-compresses in memory afterward.

Changes

• plan.md: S3.3 (full-sync uncompressed) + S3.4 (AOF uncompressed) marked complete — covered by PR . valkey-io/valkey#24. S3.1/S3.2 marked deferred, added to the §8 v2 out-of-scope list.
• design §2.6: scope note at the top of Persistence; R2.6.1–R2.6.4 tagged v2; R2.6.8 and the §1 boundary note updated for the v1/v2 split.

In v1, RDB_ENC_COMPRESSED (=4) stays reserved-but-unused and RDB_VERSION is not bumped.

Not changing

R2.6.5–R2.6.8 (AOF / replication / DUMP / full-sync all uncompressed) hold in both v1 and v2 — no behavior change.