[killertux]

The existing SliM codec used one length rule at every layer, so Unicode messages could be decoded incorrectly and malformed instructions could terminate the server. For example, nested strings were measured as UTF-8 bytes even though SliM V0.5 requires UTF-16 code units, while only the outer frame uses a byte count.

This change separates outer framing from recursive value encoding, adds a recursive SlimValue, validates frame/list structure, supports lengths beyond six digits, and preserves ordinary strings inside nested result lists. Well-formed instruction lists with unknown operations or invalid arity now return MALFORMED_INSTRUCTION for their original ID and allow the rest of the batch to continue.

The public Instruction enum gains a Malformed variant, so downstream exhaustive matches will need to handle it. Typed list instruction arguments and the typed fixture runtime remain follow-up work.

Validation:

• cargo test --workspace --all-features
• cargo clippy -p slim_protocol --all-targets -- -D warnings
• cargo fmt --all -- --check
• Independent review against the documented SliM V0.5 protocol