index: time catch-up stages and assemble heights on the script pool

[rearden-grok]

Summary

• The 10s index: build line and the one-height index: apply line report interval read= build= commit= milliseconds. tip: accept bf= stays the combined build+commit time.
• Catch-up assemble publishes one job per height to rbtc-scripts-* (filter GCS and tweak EC). The IO thread fills the block hash so those jobs do no store IO. The CPU thread still commits each window once.
• Filters and tweaks stay byte-identical. No cache, no new threads, and no write-path change.

Host check after merge

Grep index: build next= around 700k–800k and past 850k. The 2026-09-26 baseline was about 43–53/s at 711k–745k and 10–20/s from 820k to the tip.

• If build= is the long pole, rate= should move toward 30–60/s on 4 CPUs and build= should fall toward read=.
• If read= is still at least 80% of the interval past 850k, the next change is the reader (window size, ring depth), not a pipeline reshape.
• If commit= is the long pole, promote write aggregation.

Do not start an in-flight cache from this PR alone.

Test plan

• cargo test -p rbitcoin-consensus --lib index_writebehind
• Workspace suite, clippy -D warnings, deny, ast-grep, and ci-os-smoke
• Operator host catch-up log after merge

Made with Cursor