Add opt-in pipeline runtime diagnostics

[Toti330]

Summary

Adds DIAGNOSTICS=1, an opt-in end-of-run report for the generation-to-PostgreSQL pipeline.

Why this is useful

It turns a single final throughput number into measurable evidence: generated CSV payload volume, worker generation cost, worker channel backpressure, writer connection and receive time, star/planet COPY time, commit time, transaction count, and actual batch sizes. This makes it possible to distinguish CPU generation pressure, queue pressure, transfer/COPY time, and commit cost.

Implementation

• Adds inserter/src/diagnostics.rs with lock-free aggregate counters.
• Records worker generation and channel-send wait timings only when DIAGNOSTICS=1.
• Records writer connection, receive, transaction setup, star COPY, planet COPY, and commit stages.
• Uses a drop guard to report even if run() returns early after diagnostics begins.
• Documents metrics and explains why aggregate thread time can exceed wall-clock time.

Independence

This PR is based directly on main and is independently usable. It does not require precise timing, tuning, checkpoint, preflight, or progress logging changes.

Verification

• cargo test --all-targets passed in the Rust Docker toolchain.
• BENCHMARK=1 DIAGNOSTICS=1 smoke run emitted payload, generation, channel, writer, and batch diagnostics.

---

Generated by GPT-5.6 Terra & Sol. Ideas from the author; implementation by AI.

[SuperB3333]

Partially implemented in 44e7c91