fix(transport): make debug logging opt-in for message privacy

[gabrieledellamalva]

Overview

Privacy / AI agent transport hardening — fixes #786.

PostMessageTransport currently calls console.debug for outgoing and incoming JSON-RPC messages and for messages from unrelated iframe sources. This is noisy in multi-app hosts and can expose raw tool messages in developer consoles or captured browser logs even when the host never requested diagnostic logging.

Changes

• Make debug logging opt-in via an optional third constructor argument: { logger?: Pick<Console, "debug"> }.
• Keep existing two-argument constructors and message delivery compatible.
• Preserve source validation, silent rejection of unrelated/non-JSON-RPC traffic, and the existing console.error path for malformed JSON-RPC messages.
• Add regression tests covering zero default debug output and an explicitly provided logger.

Scope / security notes

This is defense-in-depth for diagnostic logging and privacy. It is not an authorization boundary, message encryption, a sanitizer for malformed-message error logs, or an MCP tool-permission firewall. Embedders that choose to enable a logger are responsible for protecting their debug output.

Validation

• Two targeted Bun regression tests added to existing transport suite.
• Fixed the logger test mock's TypeScript argument typing after a local compiler failure.
• Local macOS / Bun 1.4.2: bun test src/message-transport.test.ts — 20 passed, 0 failed.
• Local npm run build — passed.
• Local npm test — 441 passed, 1 skipped, 0 failed.
• Formatting-only follow-up commit 0936aa1 applied to both changed files; the final npm run prettier check is not independently confirmed yet.
• Upstream CI currently reports action_required (GitHub Actions permission/approval), not executed test results.

Closes #786 upon merge.