[observability] Add Python logging integration with structured events

[dgenio]

Problem

contextweaver has zero import logging across its entire codebase. Users cannot observe what the context pipeline or routing engine is doing at runtime. When a context build drops items unexpectedly or routing returns surprising results, there is no diagnostic output.

Proposal

Add Python standard library logging integration with structured messages at key pipeline stages. Zero-cost at default log level (WARNING), rich at DEBUG.

Loggers

• contextweaver.context — context pipeline events
• contextweaver.routing — routing engine events
• contextweaver.store — store operations
• contextweaver.adapters — adapter conversions

Log points (examples)

DEBUG: candidate generation counts, sensitivity filter drops, firewall interceptions, scoring top-k, dedup removals, selection packing, token budgets used

INFO: build summaries ("context build: phase=%s, included=%d, dropped=%d, tokens=%d/%d"), route summaries ("route: query=%r, top_k=%d, candidates=%s")

Sensitivity guard

• NEVER log item text content at any level
• Log only IDs, counts, scores, and structural metadata

Acceptance Criteria

• Loggers created for context, routing, store, and adapters
• DEBUG-level messages at each context pipeline stage
• INFO-level summary messages for build and route completions
• No item text content logged at any level
• Zero performance impact at default log level
• No runtime dependencies added (stdlib only)
• All existing tests still pass

File Paths

• src/contextweaver/context/manager.py, candidates.py, scoring.py, dedup.py, selection.py, firewall.py, sensitivity.py (edit)
• src/contextweaver/routing/router.py (edit)
• src/contextweaver/adapters/mcp.py, a2a.py (edit)

References

• Python logging best practices: https://docs.python.org/3/howto/logging.html