permissionDecisionReason from PreToolUse hooks not forwarded to can_use_tool callback

[Pajk]

Summary

When a PreToolUse hook returns permissionDecision: "ask" with a permissionDecisionReason, the CLI invokes the SDK's can_use_tool callback — but the reason string is not included in the wire message and is therefore inaccessible to SDK consumers.

Current wire type (SDKControlPermissionRequest)

class SDKControlPermissionRequest(TypedDict):
    subtype: Literal["can_use_tool"]
    tool_name: str
    input: dict[str, Any]
    permission_suggestions: list[Any] | None
    blocked_path: str | None
    tool_use_id: str
    agent_id: NotRequired[str]
    # ← no permissionDecisionReason

ToolPermissionContext (passed to can_use_tool) similarly has no hook-related field — only signal, suggestions, tool_use_id, and agent_id.

The only occurrence of permissionDecisionReason in the package is in PreToolUseHookSpecificOutput — an outbound hook response type, not an inbound callback field.

Use case

We have a Slack bot that uses the Agent SDK and surfaces permission requests as interactive messages. When a hook detects a dangerous command (e.g. rm -rf) and sets a warning reason, we want to display that reason to the user in the permission prompt — but it's inaccessible because the CLI doesn't forward it.

Requested change

1. Include the hook's permissionDecisionReason in the can_use_tool control message the CLI sends to the SDK
2. Add an optional field (e.g. hook_decision_reason: str | None) to SDKControlPermissionRequest
3. Expose it on ToolPermissionContext so SDK consumers can surface it in custom permission UIs