interop/v2: opt-in compressed-values profile (compressed + encrypted cross-SDK values)

[27Bslash6]

Design-discussion anchor for the interop/v2 compressed-values profile RFC (workspace ticket LAB-1135) — restoring the 2025-11-14 RFC's descoped vision of compressed AND encrypted cross-SDK values, as a versioned additive profile.

interop/v1 (#1) deliberately shipped without compression: values are bare MessagePack and the encryption AAD pins compressed = "False" (spec/interop-mode.md). That surface is frozen. The spec's own versioning rule ("versioning-by-mode-name is sufficient") sanctions a new mode: interop/v2, which wraps values in a minimal container carrying an LZ4-block-compressed (or uncompressed) payload, and pins the AAD compressed component to the frozen "True" token (#12) as a per-mode constant.

Design constraints carried in from LAB-1129 / LAB-1135:

• Deterministic AAD, no sniff-and-retry: v1/v2 discrimination is by configured mode, pre-AAD (spec/encryption.md reader rules).
• v1 compatibility frozen: every existing interop/v1 vector passes unchanged.
• msgpack bin (protocol 1.1, LAB-783: StorageEnvelope compressed_data → msgpack bin encoding (protocol 1.1) #44) is the normative byte encoding; the legacy array-of-ints leniency gets an explicit accept/reject decision.
• CRIME/BREACH verdict recorded explicitly.
• Per-SDK LZ4-block dependency bill named (the 2025 PHP fork is the cautionary tale).
• Decompression bounds are a v2-specific obligation; complementary to (not deciding) interop/v1: pin a cross-SDK collection-size bound for interop values (element caps currently SDK-local) #20's element-count axis.

Deliverable: spec/interop-v2.md + generated, independently cross-checked test vectors (tools/interop-v2-reference.py stdlib generator + tools/interop-v2-crosscheck.mjs verifier), following the interop/v1 tooling pattern. Spec PR to follow on this issue.