bug(state-manager): STATE.md timestamp time-of-day lost on unrelated edits

## Summary

During Wave 3 of switchboard-blue, the state-manager was dispatched to update two stale scalar fields (`l3_bc_count`, `l4_vp_count`) in `.factory/STATE.md`. The dispatched task touched only those two fields plus `timestamp`. The state-manager's commit replaced `timestamp: 2026-06-25T21:00:00Z` with `timestamp: 2026-06-25T00:00:00Z`, zeroing the time-of-day portion despite no part of the dispatch instructing this.

## Impact

LOW. The timestamp is informational; downstream tools likely use commit timestamps instead. But:
- Cosmetic regression every time state-manager is dispatched for narrow scalar edits.
- Suggests state-manager's "update modified date" subroutine truncates rather than uses the current time.

## Mitigation suggestions

- state-manager should write `timestamp: $(date -u +%Y-%m-%dT%H:%M:%SZ)` (full ISO-8601) rather than `${date}T00:00:00Z`.
- Or: explicitly preserve the existing time-of-day when no time change is intended.

## Reproduction

Switchboard-blue session `b8690230-7b5d-4e57-b71c-8b9ffeee3d9a`. State-manager dispatch for `F-W3-M-003` (counts fix). Resulting commit `5df5400` on `factory-artifacts`.

## Related (this batch)

- drbothen/vsdd-factory#272 — P19: architect hallucinated internal/ packages
- drbothen/vsdd-factory#273 — P20: parallel agent burst commit co-mingling
- drbothen/vsdd-factory#274 — P21: ADR amendment + code update drift across revisions
- drbothen/vsdd-factory#275 — P22: parallel agents make contradictory architectural decisions
- drbothen/vsdd-factory#276 — P23: spec-reviewer cross-eye should be mandatory post-spec-burst
- drbothen/vsdd-factory#277 — P24: derived summary fields go stale on primary content add
- drbothen/vsdd-factory#278 — P25: consistency-validator + spec-reviewer verdict gap