t27b ledger: eval_hslm and ngram_hslm are unlisted, and two not-pass slots must be freed first #10858

#10467 (b2a33e5) added specs/ml/eval_hslm.t27 and specs/ml/ngram_hslm.t27 to master with no rows in docs/reports/t27b_expectations.json. The reference passes both. t27b blocks both on ExprCall(undeclared fn): each does use ml::train_hslm; and calls into it, which is cross-module use-resolution (lane #9065).

t27b-native-ratchet therefore reports UNLISTED 2 for every PR built on master, even after #10837 lists the other eight drifted specs.

The two rows cannot just be added. The ledger sits at its cap, 139 not-pass rows (138 blocked + 1 codegen) against max_not_pass: 139, and the cap is not raised. Two blocked rows must start to pass first.

Boundary

• docs/reports/t27b_expectations.json: add both rows in LC_ALL=C sort order, and keep the not-pass count at or below 139.
• Free two not-pass slots by making t27b pass two currently blocked specs honestly, with new logic in a .t27 plan module and cli/t27b glue at net <= 0 hand-written lines. Candidates whose ledger blocker is unique:
  • specs/tri/utils/time.t27: ExprCall(@divFloor). It also uses @mod and std.os.linux.clock_gettime, so check what blocks it next before you choose it.
  • specs/tri/pipeline/builder.t27: ExprCall(@constCast) of &[_]u32{}, an empty slice.
  • Or eval_hslm and ngram_hslm themselves, by resolving use ml::train_hslm calls (t27b: resolve cross-module use names - the 28 ml specs #8993 unlocked on the reference #9065).
• Not in scope: raising max_not_pass, marking either spec n/a, deleting or weakening spec tests.

Scenarios

1. Today: a PR built on master runs t27b-native-ratchet and gets UNLISTED 2 (eval_hslm, ngram_hslm), red.
2. After the fix: the same PR gets UNLISTED 0, OVER CAP 0, UNEXPECTED PASS 0, green, with the cap unchanged.

Requirements

• Two rows: {"path": "specs/ml/eval_hslm.t27", ...} and {"path": "specs/ml/ngram_hslm.t27", ...}, blocked or pass, sorted.
• Two previously blocked rows flip to pass, each backed by a lab or CI corpus run.
• Every new decision has test blocks in its .t27 plan, and a mutant of each new branch is killed.

Criteria

• grep -o '"path": "[^"]*"' docs/reports/t27b_expectations.json | LC_ALL=C sort -c exits 0.
• Not-pass rows (blocked + codegen) <= 139, and max_not_pass unchanged or lower.
• t27b-native-ratchet is green on the PR; its verdict line shows UNLISTED 0.

Follows #10837, which freed one slot (@truncate in specs/tri/utils/utf8.t27) and listed hslm_tokenizer, hslm_corpus and six pass specs.

Success Criteria

• grep -o '"path": "[^"]*"' docs/reports/t27b_expectations.json | LC_ALL=C sort -c exits 0
• jq '[.[] | select(.status=="blocked" or .status=="codegen")] | length' docs/reports/t27b_expectations.json outputs a number ≤ 139
• jq '.max_not_pass' docs/reports/t27b_expectations.json outputs a number ≤ 139
• t27b-native-ratchet run on the PR prints a verdict line containing UNLISTED 0 and exits 0
• Two specs previously listed as blocked in docs/reports/t27b_expectations.json now show status": "pass" in the same file
• Each newly passing spec has at least one test block in its .t27 plan module and a mutant killed in CI (mutation score 100% for new branches)