fix(ci): budget complete native acceptance within 180 minutes

[afflom]

A 45-minute native bootstrap job can stop a complete consumer gate before its required application generations finish. The matrix consumer calls the public SDK eight times through ordinary build, independent verification replay, native product construction and two clean-root reproductions. Two observed full 158-vector generations took 628.66 and 650.16 seconds; a later optimized generation took 510.66 seconds. At the larger observed cost, generation alone is about 87 minutes before proofs, oracles, authority scans and planted defects.

Set the existing native bootstrap deadline to a bounded 180 minutes. Both runners, the complete just vv command, failure propagation, permissions and immutable dependency checks remain present. Add an actual-workflow regression covering the measured workload and the existing native/full-gate boundaries; the product register remains empty.

The new regression failed against 45 minutes. After the change, all 23 Rust tests, both invoked Node tests, workspace/all-target Clippy with warnings denied, formatting, model comparison and all-features/all-target checks passed through immutable SDK sha256:46cf404780a7f05fe2bc667690320198fbed1ffadf4bb1899fd0472c6171e44d. Planting 45 minutes failed the named test again; restoring the exact workflow bytes passed. These are component results, with no claim of a complete hosted gate or new installed-SDK qualification.

Policy input commit: 1c7ee5150b65bfadfe8b8b54c0e1051f68b40394. The unchanged official renderer can bind this exact revision to the new dual-native SDK when its construction finishes. This is the template dependency for lexlean-matmul PR #3 and PrismPM PR #195; it requires no additional SDK source change.