seal: re-save ml_train_hslm.json after the \xNN emitter change -- coverage gen-drift on every PR #10865

The coverage check ("Every seal still describes its spec") is red on master's PRs with one gen-drift seal:

FAIL: 1 seal(s) newly do not hold
  ml_train_hslm.json  [gen-drift]
      specs/ml/train_hslm.t27 still hashes the same, but gen_hash_rust do not

The spec is unchanged. The generated Rust changed, which is what an emitter fix looks like. The likely cause is #10812 (\xNN in a string now lowers to one byte in every backend), merged as ae19b93: train_hslm.t27 writes binary checkpoint bytes. Seen on the stack PRs #10792, #10800, #10806, #10816, #10822 and #10830, none of which touch specs/ml or the compiler (run 39198010222).

Boundary

• Re-save ml_train_hslm.json with master's t27c (t27c seal --save specs/ml/train_hslm.t27), on the Railway lab.
• Not in scope: adding the seal to seal_baseline.txt (the gate's own text says that is the wrong repair for gen-drift), or editing specs/ml/train_hslm.t27.
• The same run lists 9 baselined seals that now hold (TernaryShift.json, TriHtml.json, TriTerminal.json, TriXml.json, encoding_TriHtml.json, +4). Dropping their baseline lines is welcome in the same PR.

Scenarios

1. Today: any PR built on master runs coverage and gets FAIL: 1 seal(s) newly do not hold.
2. After the fix: the same check reports no newly failing seal.

Requirements

• The new seal's gen_hash_rust equals what master's t27c emits for specs/ml/train_hslm.t27.
• First confirm the drift comes from lexer: \xNN in a string lowers to one byte, and every backend writes the byte back #10812, by comparing the generated Rust before and after ae19b93.

Criteria

• t27c seal --verify specs/ml/train_hslm.t27 reports MATCH with master's t27c.
• coverage is green on the PR, or red only for reasons master also has.

Requirements

FR-001: Re-save ml_train_hslm.json by running t27c seal --save specs/ml/train_hslm.t27 on the Railway lab using master's t27c.
FR-002: Confirm the gen-drift originates from #10812 (ae19b93) by comparing generated Rust for specs/ml/train_hslm.t27 before and after that commit.
FR-003: The new seal's gen_hash_rust MUST equal what master's t27c emits for specs/ml/train_hslm.t27.
FR-004: Do not edit specs/ml/train_hslm.t27 and do not add the seal to seal_baseline.txt.

Success Criteria

• t27c seal --verify specs/ml/train_hslm.t27 reports MATCH with master's t27c (exit 0)
• coverage check passes on the PR (exit 0) or fails only for reasons that also fail on master
• 1 seal (ml_train_hslm.json) newly holds after the re-save
• 9 baselined seals (TernaryShift.json, TriHtml.json, TriTerminal.json, TriXml.json, encoding_TriHtml.json, +4) can have their baseline lines dropped in the same PR