fix(tool): redact permission result reasons

[XnLemon]

Objective

Close one Phase2 governance redaction gap where permission result reasons are returned to the model.

Changes

• Redact PermissionResultFor reasons for denied and approval-required tool calls.
• Redact ApprovalDeniedResultFor reasons returned after reviewer rejection.
• Cover Authorization, Cookie, api_key/token/password/secret/passwd, URL userinfo, OpenAI-style sk- tokens, JSON secret fields, and private-key blocks without adding a reverse dependency from tool to platform.
• Add regression tests for permission denial, approval-required, approval-denied, generic Authorization fallback, and multi-value Cookie headers.

Validation

• go test ./tool
• go test ./internal/flow/processor -run Permission|ApprovalDenied|ToolPermission
• go test ./tool/dynamicworkflow ./platform/toolpolicy
• go vet ./tool ./internal/flow/processor ./tool/dynamicworkflow ./platform/toolpolicy
• git diff --check
• open-code-review subagent loop: final review 0 High / 0 Medium / 0 Low

Known Risks / Limitations

• Scope is intentionally limited to permission-result reason strings returned to the model.
• A broader go test ./internal/flow/processor run on Windows still hits existing path separator assertions in skill path tests, unrelated to this change.

Follow-up

• Continue Phase2 closure sweep across remaining governance and redaction output boundaries in the next stacked increment.