own_model: ground cells 8 and 24, put the DELIVER evidence to review again #10184

Goal

Ground cells 8, 24 and 25 of specs/igla/own_model.t27, and put the evidence of the DELIVER cells (7, 8, 16, 17, 18, 25, 27) to review again. Part of epic #10024.

• Cell 8 ("A public 100M model") now has its source: playra/tern-tc-100m is public on Hugging Face.
• Cell 24 ("Heap and file I/O in t27c") now has a passing test: specs/ml/host_io_model.t27 through t27c --backend ml (t27c --backend ml: thin glue for host file I/O (Closes #10085) #10165).
• Cell 25 ("t27c --backend ml") now has its source: specs/ml/train_slice_model.t27 (spec(ml): train_slice_model -- main() trains 20 steps and writes a checkpoint to a real file (Closes #10183) #10186) compiles with --backend ml to a binary that trains one 3-head slice.
• spec(igla): own_model -- put the DELIVER cells' evidence (7, 16-18, 27) to review (Closes #10178) #10179 was meant to get the DELIVER evidence a review. An auto-merge enabled by another session merged it at 03:16Z, before the reviewer reached it, so it has none. spec(ml): train_slice_model -- main() trains 20 steps and writes a checkpoint to a real file (Closes #10183) #10186 merged at 03:42Z the same way.

Boundary

• specs/igla/own_model.t27: header entries, SOURCES[7], SOURCES[23], SOURCES[24], MISSING[7], MISSING[23], MISSING[24], GROUNDED, TEST_PASS, LIVE, STATUS, PCT, ROWS, and the tests that count them.
• specs/igla/own_model_board.t27: MSG, OVERALL_PCT, PROGRESS, LENS_PCT, CELL_STATUS, CELL_PCT, METRICS_BYTES, and the tests that count them.
• REVIEWED stays all false. Cells 9 and 26 stay gaps.

Scenarios

• Cell 8 is a DELIVER step. It is grounded and live, but has no review, so it reads DRAFT.
• Cell 24 is a PRACTICE step with a passing test, so it reads VERIFIED.
• Cell 25 is a DELIVER step. It is grounded, with no review, so it reads DRAFT; its percent (30) says the binary trains one slice, not the whole model.
• Negative control: putting STATUS[24] back to a gap makes a_gap_is_named_not_filled and a_status_is_earned_not_chosen fail.

Requirements

• FR-001: The tally MUST read "1 published, 17 verified, 7 drafts, 2 gaps; 60%".
• FR-002: t27c test-report on the Railway lab MUST give own_model 9/9 and board 6/6, with 0 fail and 0 vacuous.
• FR-003: The evidence numbers in the PR MUST come from the lab at the PR's merged head.
• FR-004: REVIEWED MUST stay all false in this change.

Success Criteria

• t27c test-report specs/igla/own_model.t27 reports 9 tests, 0 FAIL, 0 vacuous passes on the lab.
• t27c test-report specs/igla/own_model_board.t27 reports 6 tests, 0 FAIL, 0 vacuous passes.
• The specs/ml/train_slice_model.t27 binary exits 0 and writes 379 bytes to /tmp/t27_train_slice.bin.
• With STATUS[24] set back to a gap, own_model reports 2 FAIL.

Close criteria

• The PR merges with both specs green on the lab.
• It has an approving review that a later PR can name as the REVIEW_ID of the DELIVER cells.