credential_redactor: docs imply redact() scrubs PII, but it covers secrets only; SSN patterns diverge

Found during a code-quality / docs pass.

In `agent_os/credential_redactor.py` (around lines 101-199), `redact()` / `redact_data_structure()` iterate only `PATTERNS` (secret-like material), not `PII_PATTERNS` (email / phone / SSN / credit-card / IP). So PII is *detected* by `find_pii_matches()` but **not removed** by `redact()`. The class docstring's 'redact sensitive material' phrasing can lead a caller to assume PII is scrubbed when persisting/returning data.

Suggestion: either add an opt-in `redact_pii=True` path, or clarify in the docstring that `redact()` is secrets-only.

Separately, the SSN pattern here (`\d{3}-\d{2}-\d{4}`) is narrower than the one in `integrations/base.py` (which also accepts space/dot/no-separator) — reconciling to one shared pattern would avoid divergent detection.

Start in agent_os/credential_redactor.py around lines 101-199, comparing redact(), redact_data_structure(), PATTERNS, and PII_PATTERNS; then compare the SSN pattern with integrations/base.py. Confirm whether the intended fix is an opt-in PII path or secrets-only documentation, and reconcile the SSN behavior if required; done means the documented and actual redaction and detection scopes agree.

Written by the indexing model from the issue text.