Add logging instrumentation to 5 uninstrumented source modules

Factory experiment 1. Hypothesis: Add logging instrumentation to 5 uninstrumented source modules to improve observability score.

What to Build

Add use crate::logging; and targeted logging::info/error/warn/debug calls to these 5 files:

1. src/channel.rs (26 functions) — log channel send/receive operations, errors, connection state changes
2. src/process_memory.rs (31 functions) — log memory snapshots, allocation events, cleanup operations
3. src/telemetry.rs (66 functions) — log telemetry event emission, queue state, flush operations
4. src/message.rs (7 functions) — log message processing, format conversions, error paths
5. src/embedding_stub.rs (15 functions) — log embedding lookup requests, cache hits/misses, fallback behavior

Add 3-5 meaningful log calls per file at:

• Function entry points for key operations
• Error paths (especially where errors are silently swallowed)
• State transitions

Acceptance Criteria

• All 5 files have use crate::logging; imports
• At least 15 new logging calls total (3+ per file)
• Log calls use appropriate levels (info for normal ops, warn for unusual, error for failures, debug for verbose)
• Tests pass
• Eval score does not regress

Constraints

• Read CLAUDE.md before starting
• Do NOT touch files outside declared scope (src//*.rs, crates//*.rs)
• Do NOT modify eval/score.py or .factory/ files
• Use the existing crate::logging module (info, error, warn, debug functions)
• Do NOT add new dependencies

Read CLAUDE.md first, then inspect the existing crate::logging module and the five named files: src/channel.rs, src/process_memory.rs, src/telemetry.rs, src/message.rs, and src/embedding_stub.rs. Add the specified imports and meaningful calls within those files, then run the tests and confirm at least 15 calls are present without eval score regression.