14654 Add Redshift Data Connections driver

[softwarenerd]

Summary

Adds a Redshift data connection driver (positron-data-driver-redshift). Redshift speaks the PostgreSQL wire protocol, so the driver is built on pg and shares structure with the PostgreSQL driver, with additional handling for the ways a serverless warehouse differs from a local Postgres: cold-start connects, idle-dropped sockets, and a per-statement latency floor.

What's included

• Redshift driver: user/password connection (host accepts a full host:port/database endpoint paste), SSL on by default, schema/table/view/column browsing, and generated connection code for redshift_connector (Python) and DBI/RPostgres (R).
• Cross-database browsing: on RA3/Serverless clusters the tree gains a top-level Databases group and browses every database in the namespace over one connection via the SVV_ALL_* catalog views and three-part references, detected at runtime.
• Connection resilience (also applied to the PostgreSQL driver): TCP keepalive so idle sockets survive, an idle-error guard so a dropped socket can't crash the extension host, transparent reconnect-and-retry when a query hits a dead connection, and bounded connect-retry with backoff so a resuming serverless workgroup is waited out instead of failing.
• Data Explorer summary performance: column profiling is computed in ~3 batched statements per request (one scalar-aggregate scan covering null counts, summary stats, and an exact PERCENTILE_CONT median; one UNION ALL for all histograms; one UNION ALL for all frequency tables) instead of the previous ~4-5 statements per column. Redundant per-column null-count and quantile round-trips are eliminated.
• Pass coalescing: the summary panel re-requests profiles on layout churn, so the RPC handler now runs at most one pass per dataset at a time and cancels superseded passes at statement boundaries, so a burst of requests can't stack statements on the single connection.
• Type-safe value rendering and graceful degradation: frequency-table values are rendered to text per Redshift type (booleans via CASE, SUPER via JSON_SERIALIZE, VARBYTE via TO_HEX, spatial via ST_AsText, else cast), and the histogram/frequency batches degrade to a missing sparkline rather than failing the whole pass if a value type still can't be rendered.

Why the profiling rework

On the test cluster (8 RPUs, serverless), every statement carries roughly a 1-second fixed cost regardless of data size, so the original design (a separate scan per column per metric, run serially over a single connection) took 60+ seconds and timed out even on tiny tables (a 370K-row, 19-column table; a 365-row, 8-column table). The bottleneck was statement count and per-statement overhead, not scan time, so the fix was to collapse each request to a handful of set-based statements and stop redundant/overlapping work rather than to sample or approximate. The frontend's windowed chunking (PROFILE_CHUNK_SIZE) is left unchanged and honored.

Testing

• Unit tests for the driver (connection lifecycle, cross-database detection, schema browsing, endpoint parsing), the reconnecting client (keepalive config, reconnect-and-retry, coalesced reconnect, connect-retry vs terminal-error), and the table view (batched scalar/histogram/frequency statements, boolean rendering, graceful degradation).
• Manually validated against a live Redshift Serverless workgroup: connect through a cold start, browse single-database and cross-database trees, and open multiple tables in the Data Explorer with the summary sidebar. A diagnostic "Redshift Data Explorer" output channel logs the per-pass query timeline.

Known limitations / follow-ups

• The equivalent Data Explorer profiling rework has not yet been ported to the PostgreSQL driver (its per-statement floor is far lower, so it is less urgent).
• Frequency tables are batched per chunk, so an unrenderable column type drops that chunk's frequency tables (not just the one column) rather than isolating to the single column; the summary still renders.
• SSL currently connects without server-certificate verification (the Redshift CA is not yet bundled); IAM and Okta auth mechanisms are stubbed for later.

[github-actions]

E2E Tests 🚀
This PR will run tests tagged with: @:critical @:connections

Why these tags?

Tag	Source
@:critical	Always runs (required)
@:connections	Changed files

More on automatic tags from changed files.

Warning

This PR touches a Positron directory that isn't mapped in test-tag-paths-map.json: extensions/positron-data-driver-redshift/. Add an entry (an e2e tag or [] for no coverage) so future changes are tagged automatically.

^{readme  valid tags}