fix(optimizer): skip converged PH iterations

[frahlg]

What changed

• stop progressive hedging before its iteration loop when the initial scenario decisions already meet the residual tolerance;
• report ph_iterations: 0 for that path;
• add a deterministic one-scenario regression that fails if a second solver call occurs.

Root cause

The optimizer calculated the initial residual, but checked it only after one augmented PH iteration. A one-scenario plan has no non-anticipativity conflict, so its residual is already zero. The extra solve could hit USER_LIMIT and discard a valid plan.

Measured before the fix, the one-scenario repro made two solver calls and returned internal_error with USER_LIMIT in 5 of 5 runs. After the fix it makes one solver call, reports zero iterations and returns a replay-valid plan.

Checks

• focused PH tests: 4 passed;
• focused regression repeat: 20 of 20 passed;
• full optimizer suite: 167 passed;
• make verify-all passed, including Go tests, Core to optimizer contracts, vet, builds and release-target cross-builds;
• two independent read-only reviews found no P0-P3 issue.