t27b: a signed @mod of any width up to i64 widens to the i64 routine (arena_heap, file_io_contract, reel_particles_mux) #10423

Follow-up to #10405 / PR #10410, t27b coverage epic #6063.

Boundary

• specs/tri/t27b/libm_plan.t27 (the decision) and its generated gen/rust/tri/t27b/libm_plan.rs (t27c gen-rust).
• specs/tri/t27b/conformance/rem_int.t27 (conformance cases; no new file).
• cli/t27b/src/lower.rs libm_call glue only, net <= 0 lines.
• docs/reports/t27b_expectations.json rows and .trinity/seals/ for the specs that move.
• Out of scope: a float @mod, a signed @mod wider than i64 (no i128 routine), any other builtin.

User Scenarios

• A spec author writes @mod(n, 2) on an i32 (specs/ml/arena_heap.t27 is_pow2), @mod(size, 4) (specs/ml/file_io_contract.t27) or @mod(ms, 1000) (specs/t27f/reel_particles_mux.t27 millis). Today t27b refuses each file with ExprCall(@mod): "a signed @mod is ported for i64 only" (REFUSE_WIDTH). After this, t27b compiles them and gives the reference's verdict.

Requirements

• A signed @mod narrower than i64 (i32, i16, i8, odd widths) lowers by widening both operands (the second takes the first one's type) to i64, calling the existing t27b_libm_mod, and narrowing the result back. This is lossless because the result is 0 <= r < b and b fits the operand type, and a zero or negative divisor traps or panics as before.
• The choice (widen and narrow rather than one routine per width) is recorded in libm_plan.t27 with tests.
• A signed @mod wider than i64 stays refused with REFUSE_WIDTH. An integer @rem of every width stays t27b's %.

Success Criteria

1. t27b test --blockers no longer lists ExprCall(@mod) for specs/ml/arena_heap.t27, specs/ml/file_io_contract.t27 or specs/t27f/reel_particles_mux.t27.
2. t27b test --check agrees with t27c test-report on rem_int.t27 and libm_plan.t27.
3. The ledger rows in docs/reports/t27b_expectations.json are updated for any spec that moves to pass.
4. Seals are re-saved, and the hand-written lower.rs is net <= 0 lines.

Requirements

FR-001: A signed @mod narrower than i64 (i32, i16, i8, odd widths) MUST widen both operands (the second operand takes the first operand's type) to i64, call the existing t27b_libm_mod, and narrow the result back to the original width.

FR-002: The widening-and-narrowing design choice MUST be recorded in specs/tri/t27b/libm_plan.t27 with accompanying tests.

FR-003: A signed @mod wider than i64 MUST continue to be refused with REFUSE_WIDTH.

FR-004: An integer @rem of every width MUST continue to lower to t27b's % operator.