docs: clarify Application Insights ingestion, cold_start semantics, and KQL shapes

[yeongseon]

Closes #79

Summary

P2 documentation accuracy fixes from the external review. Pure docs change — no source modifications.

1. Application Insights scope disclaimer (early in README)

Added a Scope disclaimer callout immediately after the What it does bullet list, before any KQL example. The previous "Application Insights-ready" framing was technically accurate but implicit — the disclaimer existed only midway through the README, after readers had already seen examples assuming customDimensions parsing.

The new disclaimer states plainly: this package writes structured JSON to Python logging / stdout. How those fields appear in Application Insights depends on the Azure Functions host, worker, logging configuration, and ingestion pipeline. The library does not own ingestion or schema mapping. Both customDimensions-parsed and raw-message shapes are valid in production.

2. cold_start semantics documented precisely

Added a callout in both README (Invocation Context section) and docs/usage.md (Cold Start Detection section):

cold_start=True means the first invocation observed by this Python worker process after module load. It is not a platform-level cold start metric and does not correspond to App Service plan / instance allocation cold starts reported by Azure Functions metrics.

This matches the actual implementation: _cold_start is a process-global flag flipped on the first inject_context() call after module import. It is informational for log correlation, not a substitute for Azure metrics.

3. KQL examples split into both ingestion shapes

The README Query in Application Insights section now has two clearly labeled subsections:

• When JSON fields are parsed into customDimensions — the original queries.
• When JSON remains in the message column — same queries rewritten with parse_json(message) and tostring(payload.<field>).

Both shapes occur in real Azure Functions deployments depending on the worker, host configuration, and Application Insights connection mode. Documenting only the parsed shape misled users whose payloads land as raw text.

Validation

• make lint — clean (no source files touched, but README/docs changes verified through pre-commit hooks)
• make test — 142 passed, 3 skipped, 97.05% coverage
• No source changes; no behavior changes; no public API changes.

Acceptance checklist (from #79)

• README "Application Insights-ready" disclaimer added near the top
• cold_start semantics documented in README and docs/usage.md
• KQL examples split into both ingestion shapes
• No code changes; pure docs PR