Add structlog to uninstrumented high-traffic modules

[abhi1092]

Factory experiment 4. Hypothesis: Add structlog to 4 uninstrumented high-traffic modules.

What to Build

Add structured logging via structlog.get_logger() to these modules:

1. factory/mcp_server.py (8 functions, 0 log statements): Add log = structlog.get_logger() and log at function entry for list_tools, call_tool, server startup/shutdown, and tool dispatch errors.

2. factory/runners/__init__.py (2 functions, 0 log statements): Log runner selection and initialization.

3. factory/runners/_stream.py (3 functions, 0 log statements): Log stream processing start, chunk handling, and completion.

4. factory/runners/protocol.py (2 functions, 0 log statements): Log protocol method dispatch.

Use log.info() for key operations, log.debug() for routine paths, log.warning() for fallback/error paths. Follow existing codebase patterns — bind context with log.bind(runner=..., role=...) where relevant.

Acceptance Criteria

• All 4 modules have structlog logging
• At least 15 new log statements across the modules
• Follow existing logging patterns (structlog.get_logger() at module level)
• All existing tests pass
• Eval score does not regress

Constraints

• Read CLAUDE.md before starting
• Do NOT touch files outside declared scope
• Do NOT modify eval/score.py or .factory/
• Do NOT add logging to pure data models (models.py, ace/models.py)
• Use log.debug() for routine operations, log.info() for key state transitions