Add a dormant, best-effort v4 RDS Postgres dual-write to the CI benchmark emitters

[connortsui20]

The CI benchmark emitters currently need a manual vortex-bench-migrate run to populate the new v4 RDS Postgres database. This adds a best-effort, env-gated dual-write so the emitters populate it directly, ported from the unmerged ct/bench-v4 branch onto current develop. It ships dormant: every v4 step is gated on vars.GH_BENCH_INGEST_ROLE_ARN being set and carries continue-on-error: true, and the existing v2 (static S3 JSON) and v3 (DuckDB + /api/ingest) paths are left fully intact and run first. With the variable unset the new code is dead-but-safe — a v4 write failure can never fail a workflow. Turning it on (setting the role ARN, repointing the site URL) is a separate ops change after this merges.

What's in the diff:

• A Python port of the measurement_id xxhash64 key (scripts/_measurement_id.py), the cross-language golden vectors, and a parity test asserting byte-for-byte agreement with the Rust reference. The golden vectors are the contract.
• A --postgres mode in scripts/post-ingest.py: RDS IAM auth, verify-full TLS, a five-table-plus-commit upsert in a single transaction, a NaN/Inf guard, and a best-effort site revalidate. The existing --server (v3) path stays stdlib-only and untouched.
• Testcontainer writer tests against Postgres 16, a revalidate test, and an operator cross-check utility.
• The CI wiring (parity test in the required job, testcontainer suite in a docker-gated job) and the dormant v4 ingest step added to the three emitter workflows.

The parity test and the Postgres-16 testcontainer suite run in CI. This branch also carries big-plans orchestration scaffolding under .big-plans/ (the spine, per-step plan files, and the plan: commits); that's transient and will be removed in a follow-up cleanup PR rather than being part of the shipped feature. A few hardening and doc nits surfaced during review are tracked as deferred, non-blocking items in the spine.

🤖 Generated with Claude Code

[codspeed-hq]

Merging this PR will degrade performance by 10.28%

⚠️ Unknown Walltime execution environment detected

Using the Walltime instrument on standard Hosted Runners will lead to inconsistent data.

For the most accurate results, we recommend using CodSpeed Macro Runners: bare-metal machines fine-tuned for performance measurement consistency.

⚡ 3 improved benchmarks
❌ 8 regressed benchmarks
✅ 1570 untouched benchmarks

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	Simulation	decompress_rd[f64, (10000, 0.01)]	108.9 µs	139.3 µs	-21.82%
❌	Simulation	decompress_rd[f64, (10000, 0.1)]	109.2 µs	139.6 µs	-21.79%
❌	Simulation	decompress_rd[f64, (10000, 0.0)]	108.9 µs	139.2 µs	-21.77%
❌	Simulation	decompress_rd[f32, (100000, 0.0)]	496.1 µs	583.9 µs	-15.04%
❌	WallTime	cuda/bitpacked_u8/unpack/3bw[100M]	299.8 µs	352.7 µs	-15%
❌	Simulation	decompress_rd[f32, (10000, 0.1)]	78.2 µs	91.4 µs	-14.4%
❌	Simulation	decompress_rd[f32, (10000, 0.01)]	78.2 µs	91.1 µs	-14.16%
❌	Simulation	decompress_rd[f32, (10000, 0.0)]	78.7 µs	91.3 µs	-13.88%
⚡	Simulation	bitwise_not_vortex_buffer_mut[128]	244.4 ns	215.3 ns	+13.55%
⚡	Simulation	bitwise_not_vortex_buffer_mut[1024]	304.7 ns	275.6 ns	+10.58%
⚡	Simulation	eq_i64_constant	317.9 µs	287.9 µs	+10.42%

Tip

Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing ct/bench-v4-emitters (9e601fb) with develop (9814173)}

[connortsui20]

Superseded by #8513, a minimal code-only PR: just the --postgres writer, the measurement_id port, and the dormant workflow steps. The big-plans orchestration scaffolding and the test suite (testcontainer, revalidate, golden parity, cross-check) are intentionally left out of the repo.