Prove O(n) fixed-accuracy Hopf T-depth with charged unary phase sources

[GoGoKo699]

The previous fixed-accuracy complete-frame schedule had O(n log log n) T-depth because its geometric source reflections were repeated within each group. This update proves a different charged unary phase-source construction with O_eta(n) T-depth, retaining O_eta(sqrt N) T-count, O_eta(N) Clifford count, two external clean flags, and sufficient C_eta sqrt N arbitrary dirty workspace.

The proof includes the guarded Karatsuba bilinear XOR, constant-depth programmed cyclic shifts, coherent signed one-hot selection, native source preparation and actual inversion, full inactive-sector identity, and a 2 delta source-return bound per group. A natural suffix cutoff plus capped-precision chunked tail closes the same-circuit global ledger.

Six bounded tests cover exact rank/shift identities, the native guard and small source preparations, unequal three-stage rows, and full source-return error. They are not a scalable native QROM/frame emitter. Current-frontier documents and source attribution are updated, including catalytic/shared-phase prior art.

Validation: independent local/global/skeptical audits found no blockers. All 466 tests pass locally and on Python 3.11/3.13. All four exact-receipt suites, reviewer walkthrough, bounded native examples, and resource-ledger checks pass. All five CI checks are green on head ae42701, including desktop/narrow rendered presentation after a display-equation formatting fix.

The available depth lower bound remains Omega(1). No depth-optimality, generic constant-depth rotation, practical crossover, or high-precision endpoint claim is made.