Add canonical logic derivation task

[sileod]

What changed

• add a logic_derivation task reusing logic_depth theory generation, rendering, focusing, metadata, and difficulty controls
• add bounded proof-term dynamic programming with the objective: minimum derivation depth, then minimum proof-tree size
• reject examples with multiple optimal proof strings
• preserve generation balancing state only for accepted examples
• add tests for size tie-breaking, ambiguity rejection, registration, generation, and scoring

Answer format

Facts are line numbers, and rule applications are recursive terms, for example 8(2,7(0,3)).

Validation

• Python syntax compilation for both new files
• isolated execution tests for unique proofs, ambiguous proofs, tie-breaking, and zero-body rule serialization

The full repository test suite was not run locally because this environment has no repository checkout or outbound GitHub access.