Fix versioned entry witness placement ABI

[a19q3]

Summary

• introduce the versioned cellscript-witnessargs-input-type-v2 placement contract for generated CKB entry wrappers
• load the entry witness from group input 0, with group output 0 as the output-only type-group fallback
• preserve raw CSARGv1 compatibility while strictly decoding canonical three-field WitnessArgs and selecting only input_type
• update the CKB adapter, compiler metadata, CLI output, ABI documentation, and backend shape baselines

Problem

Generated action entry wrappers previously loaded the raw witness at transaction-global input 0. That is incorrect for script groups whose first input is not input 0, and it collides with canonical CKB lock witnesses because a signing lock owns WitnessArgs.lock.

ABI and compatibility

For the CKB target profile, a versioned entry payload is placed in WitnessArgs.input_type. The generated wrapper uses group input 0 and falls back to group output 0 only for output-only type groups. It never aliases output_type. Raw CSARGv1 remains accepted for compatibility. Malformed or non-canonical WitnessArgs fails closed with entry decode error 25.

The adapter placement API preserves an existing lock field and refuses an already occupied input_type field.

Verification

• six real CKB-VM regressions cover canonical multisig lock coexistence, a type group beginning at global input 1, raw-v1 compatibility, output-only fallback, output-type non-aliasing, and malformed field rejection
• ./scripts/cellscript_gate.sh dev
• ./scripts/cellscript_gate.sh ci
• pinned backend gate against CKB revision f7fa443 with direct node acceptance
• downstream Myelin production gate, including live parent CKB devnet submission with canonical multisig and negative tamper/conflict probes

The generated token-transfer ELF remains within the existing 7,168-byte backend budget.

Downstream integration: Myelin-Labs/Myelin@7deb8e9