docs(features): add Database Replication and CDC, and the Analytics Replica use case

[claudespice]

Summary

Database replication had no feature page, despite CDC being the capability behind refresh_mode: changes. It was only findable as a FAQ answer and a handful of scattered mentions. This adds the missing feature page and the use-case page for the architecture built on it.

New: features/database-replication-and-cdc.md — what database replication and change data capture are, why it beats ETL / read replicas / HTAP for analytical load, the supported sources and their mechanisms, configuration, PostgreSQL source prerequisites, and how hard deletes differ from incremental ingestion.

New: use-cases/analytics-replica.md — the analytics replica pattern: attach a columnar node to an operational database, load a snapshot, then apply committed changes continuously so analytics runs seconds behind without touching the primary. Covers why it differs from a read replica (storage layout and execution engine, not just load placement), table-by-table adoption, and governing access at the replica.

Both are wired into the root SUMMARY.md, and the Welcome page's use-case table gains an Analytics Replica row. features/data-acceleration/README.md gains one line pointing refresh_mode: changes at the new page.

Pages updated

• features/database-replication-and-cdc.md — new
• use-cases/analytics-replica.md — new
• SUMMARY.md — Database Replication and CDC under Features, Analytics Replica under Use-Cases
• README.md — one row in the use-case table
• features/data-acceleration/README.md — cross-link from the refresh-modes table

Verification

• Live-ness evidence: every capability documented is already shipped and publicly described. Replication via refresh_mode: changes is documented in the public OSS docs, and native PostgreSQL WAL, MongoDB Change Streams, and DynamoDB Streams CDC all appear in shipped releases in changelog/README.md.
• Source list verified against the public OSS documentation, not the launch blog. The blog announcing this pattern lists MySQL among the sources; the OSS docs state MySQL has no native change feed and requires Debezium. The page follows the docs — MySQL and SQL Server are listed under Debezium. No forward-looking availability is stated anywhere.
• PostgreSQL prerequisites (wal_level = logical, REPLICA IDENTITY FULL, the REPLICATION role attribute) and the acceleration block taken from the published OSS reference.
• Both new pages appear in the root SUMMARY.md, so neither ships unreachable.
• Internal links resolve. The five pre-existing broken links reported in README.md are untouched by this branch — see below.
• GitBook {% hint %} syntax, not :::. building-blocks/ untouched.
• No private-repo references in branch name, commits, title, body, or content.

Coordination with #127

#127 (open, sitewide link repair) touches README.md and faq.md. This branch was scoped to stay out of its way:

• The README.md change here is a single new row in the use-case table, in a region fix(docs): repair broken and cross-space docs links sitewide #127 does not touch. The five pre-existing broken links in that file are all ones fix(docs): repair broken and cross-space docs links sitewide #127 fixes, so they were deliberately left alone.
• faq.md was not edited even though its CDC answer is the natural place to link this from, because fix(docs): repair broken and cross-space docs links sitewide #127 rewrites those exact lines. Worth a follow-up once it merges: the answer omits MongoDB Change Streams, which is a native CDC source.

Not a blocker for this PR, but worth a look before #127 merges: it appears to repoint the Apache Kafka bullet in faq.md at data-connectors/debezium.md instead of data-connectors/kafka.md.

Update — page renamed

Renamed to Database Replication and CDC (from "Replication") at review request. The file was renamed to features/database-replication-and-cdc.md to match, along with the SUMMARY.md entry and both inbound cross-links.

Renaming the file was free here because the page has never been published — doing it after merge would have needed a GitBook redirect. The body now introduces "change data capture (CDC)" explicitly as the mechanism, rather than only describing change feeds.

origin/trunk was merged into the branch (no force-push) to pick up the apps→projects portal sweep that landed while this was open.