feat: enable writable hook properties for tool and result modification

[github-actions]

Motivation

The Python SDK allows hook callbacks to modify event properties to change agent behavior (e.g., switching tools, modifying parameters, transforming results), but the TypeScript SDK marked these properties as readonly. This prevented users from implementing similar behavior modification patterns and created an API inconsistency between the two SDKs.

Resolves: #14

Public API Changes

The following hook event properties are now writable, allowing hook callbacks to modify agent behavior:

BeforeToolCallEvent:

// Before: Cannot modify
event.tool = differentTool        // TypeScript error
event.toolUse = modifiedParams    // TypeScript error

// After: Can modify to change behavior
event.tool = differentTool        // ✓ Agent uses different tool
event.toolUse = {
  ...event.toolUse,
  input: modifiedInput            // ✓ Agent uses modified parameters
}

AfterToolCallEvent:

// Before: Cannot modify
event.result = modifiedResult     // TypeScript error

// After: Can modify to transform results
event.result = new ToolResultBlock({
  toolUseId: event.result.toolUseId,
  status: 'success',
  content: [new TextBlock('Modified result')]  // ✓ Agent returns modified result
})

The changes are backward compatible. Existing hooks that only read properties continue to work without modification.

Use Cases

• Tool routing: Intercept tool calls and redirect to alternative implementations (e.g., sandbox, mock, or different version)
• Parameter validation/transformation: Modify tool inputs before execution (e.g., sanitization, schema migration, default injection)
• Result transformation: Post-process tool results (e.g., filtering sensitive data, format conversion, error recovery)
• Monitoring: Change tool execution based on runtime conditions without modifying tool implementations

Implementation Notes

The agent now reads potentially modified values after yielding hook events:

1. Yield BeforeToolCallEvent and capture modified tool and toolUse
2. Execute tool using potentially modified values
3. Yield AfterToolCallEvent and capture modified result
4. Return potentially modified result to agent loop

This follows the same pattern as the existing AfterModelCallEvent.retryModelCall property.