t27o: Zig front end step 2 -- std.math.pow, imported consts, --check on .zig, sweep lists .zig (from step 1 measurements) #10823

Goal

Step 2 of the Zig front end (#10597), from what step 1 (#10765, PR #10819) measured: the remaining refusals and the gaps around tri t27o <file>.zig. Option B of #10522.

Measured by step 1 (trinity 0770abb99 src/, trios 483dd9ee crates/)

2,573 .zig files, 353 refused as generated; 1,839 fns convert (1,356 distinct bodies); 606 specs carry translated tests, 590 pass all of them under t27b test (1,732 tests), 9 fail 17 tests (the same Zig fails under zig test), 7 are blocked. Tier-1-like distinct bodies with a same-file test: 878 of 1,435 convert (61 %; census target about 1,052).

Deliverable

• std.math.pow on f64 -- the largest remaining refusal (103): a t27 routine whose results equal Zig's on recorded vectors, or a stated refusal kept.
• Consts imported from another file (@import("x.zig").C): follow the import within the same repository when the const is a scalar literal.
• tri t27o <file>.zig --check / --oracle fail today with "rustc refused the oracle harness" (they run the Rust path): for Zig, --check writes the spec under specs/t27o_check/ and runs t27b test on it (the translated tests ARE the oracle); --oracle says it does not apply.
• tri t27o-sweep lists .zig beside .rs (sweep_path in ledger.t27), so Zig coverage is re-measured in seconds like Rust.
• The 9 failing specs: report each to its repository as a finding (the Zig itself fails its own test), not a converter bug.

Boundary

• May touch: specs/tools/t27o/zig_convert.t27, specs/tools/t27o/ledger.t27, their gen files and seals, cli/tri/src/main.rs (net-zero hand lines).
• Must not touch: rust_convert.t27, spice_convert.t27.

Links

#10597, #10765, #10522.

User Scenarios

• Given a Zig file that calls std.math.pow on f64, When tri t27o converts it, Then the generated spec contains a routine that produces bit-identical results to Zig's std.math.pow on the recorded test vectors.
• Given a Zig file that uses @import("other.zig").CONST where CONST is a scalar literal, When tri t27o converts it, Then the generated spec inlines the literal value from the imported file.
• Given a Zig file, When running tri t27o file.zig --check, Then the command writes the spec under specs/t27o_check/ and runs t27b test on it, exiting 0 if all translated tests pass.
• Given a Zig file, When running tri t27o file.zig --oracle, Then the command prints that --oracle does not apply to Zig and exits 0.
• Given a repository with Zig files, When running tri t27o-sweep, Then the output lists .zig files alongside .rs files with coverage metrics.

Requirements

FR-001: The converter MUST handle std.math.pow on f64 by generating a t27 routine whose results match Zig's implementation on recorded test vectors, or emit a documented refusal kept in the ledger.
FR-002: The converter MUST resolve @import("x.zig").C when C is a scalar literal (integer, float, boolean, or string) within the same repository and inline the literal value in the generated spec.
FR-003: The CLI MUST implement tri t27o <file>.zig --check to write the converted spec under specs/t27o_check/ and execute t27b test on it, returning the test runner's exit code.
FR-004: The CLI MUST implement tri t27o <file>.zig --oracle to print a message stating that --oracle does not apply to Zig front end and exit 0.
FR-005: The sweep command MUST list .zig files in its output via sweep_path in ledger.t27, measuring Zig coverage analogously to Rust.
FR-006: The 9 failing specs identified in step 1 MUST be reported as findings to their respective upstream repositories (since Zig's own tests fail), not tracked as converter bugs.

Success Criteria

• tri t27o specs/tools/t27o/zig_convert.t27 --check exits 0 and t27b test specs/t27o_check/zig_convert.t27 reports 0 failures.
• tri t27o fixtures/zig/pow_test.zig --check exits 0; the generated spec contains a routine for std.math.pow f64 that passes 103 recorded vector tests.
• tri t27o fixtures/zig/import_const.zig --check exits 0; the generated spec inlines the imported scalar literal value.
• tri t27o fixtures/zig/sample.zig --oracle prints "oracle does not apply to Zig front end" and exits 0.
• tri t27o-sweep lists at least 2,500 .zig files with coverage metrics in the same format as .rs files.
• All 9 previously failing specs are documented as upstream findings with repository links in specs/tools/t27o/ledger.t27.