duplicate-bodies red on master: train_slice_model copies file_io_contract's byte_sum #10203

Boundary

• specs/ml/file_io_contract.t27: byte_sum becomes pub
• specs/ml/train_slice_model.t27: its private copy byte_total is deleted and the two call sites use byte_sum

User Scenarios

• Since 0d6c3d3 (spec(ml): train_slice_model -- main() trains 20 steps and writes a checkpoint to a real file (Closes #10183) #10186), the duplicate-bodies check fails on master and on every open PR: byte_total in train_slice_model.t27 is a line-for-line copy of byte_sum in file_io_contract.t27, which that spec already imports (use ml::file_io_contract;).
• After the fix there is one checksum body, and a bug in it is fixed once.

Requirements

• No behaviour change: the two bodies are identical, so the checksum of every image stays the same.
• No ledger bless: the copy is removed, not recorded.

Success Criteria

• python3 tools/dupe_scan.py exits 0 ("no new duplicate body").
• t27c check and t27c test-report pass for both specs, and for specs/ml/host_io_model.t27, which also imports file_io_contract.
• The duplicate-bodies check is green on the PR.

Requirements

FR-001: The byte_sum function in specs/ml/file_io_contract.t27 MUST be made public so that importing specs can call it directly.
FR-002: The private byte_total function in specs/ml/train_slice_model.t27 MUST be deleted and its two call sites MUST be updated to use ml::file_io_contract::byte_sum instead.
FR-003: The checksum behavior for every image MUST remain identical to before the change (no behavioral change).
FR-004: The duplicate function body MUST be removed from the codebase rather than recorded in any ledger or allow-list.