spec(ml): train_slice_model -- reuse file_io_contract's byte_sum instead of a copy (dup-body ratchet red on master) #10206

Goal

Turn the Duplicate Body Ratchet on master green again. #10186 added byte_total to specs/ml/train_slice_model.t27, a copy of byte_sum in specs/ml/file_io_contract.t27. The ratchet now fails on master (f581a2c) and on every PR: "a body written under byte_sum is now copied 2 time(s) and was in no ledger". Part of epic #10024 (cell 25).

Boundary

• specs/ml/file_io_contract.t27: byte_sum becomes pub; its body does not change.
• specs/ml/train_slice_model.t27: byte_total is deleted and its two call sites call byte_sum, which the spec already reaches through use ml::file_io_contract.
• No ledger is blessed: the copy is removed, not declared.

Scenarios

• Given the change, When tools/dupe_scan.py runs, Then it reports no new group for byte_sum.
• Given the change, When the train_slice_model binary runs, Then it writes the same 379-byte checkpoint as before (sha256 0dfdef22...cdbeb), because the checksum function is the same body.

Requirements

• FR-001: byte_total MUST be gone from specs/ml/train_slice_model.t27.
• FR-002: specs/ml/file_io_contract.t27 MUST still pass 12/12 and specs/ml/train_slice_model.t27 5/5, 0 vacuous, on the Railway lab.
• FR-003: No duplicate-body ledger file may change.

Success Criteria

• python3 tools/dupe_scan.py exits 0.
• The specs/ml/train_slice_model.t27 binary exits 0 and its /tmp/t27_train_slice.bin is 379 bytes with sha256 0dfdef2200b3daa2930cb081f8fb4a58caf5677bae3f329ee23f0e2c328cdbeb.

Close criteria

• The PR merges with the Duplicate Body Ratchet green.