tech-debt: normalize ~196 inline datetime.now(utc).isoformat() call sites onto utc_timestamp() (style sweep) #12726

Part of the #12645 fork-convergence drive — tracked SEPARATELY because it is a large style-normalization sweep, not a fork-convergence collapse.

C1 — inline datetime.now(timezone.utc).isoformat() → utc_timestamp()

• Canonical: autobot_shared/time_utils.py:49 (utc_timestamp() is literally datetime.now(timezone.utc).isoformat()).
• Scope: ~196 files across autobot-backend/ and autobot-slm-backend/ use the inline expression instead of the shared helper (e.g. phase_progression_manager.py, takeover_manager.py, diagnostics.py, auth_middleware.py, slm/agent/agent.py, monitoring/performance_monitor.py, …).
• These are inline expressions, not forked helper copies, so this is a maintainability/consistency sweep, not a duplicate-behavior collapse. Behavior is identical (the helper is the same expression).

Why separate / lower priority

• Large blast radius (~196 files) → must be done as its own dedicated, chunked multi-PR sweep with careful diff review, not lumped into the small mechanical batches.
• Zero behavior change, so purely a consistency win — lower priority than genuine duplication.

Suggested approach (if pursued)

• AST/libcst codemod (not regex) to replace the exact inline call, add the import, verify no shadowing; split by top-level package into reviewable PRs; startup-import-smoke + unit tests per chunk.

Decision needed

Owner: pursue this sweep now, defer, or WONTFIX (accept inline usage as acceptable)?