docs: sweep versions to the 0.23.0 train

[fupelaqu]

Version sweep for the 0.23.0 train: core/repl 0.23.0 · extensions 0.3.2 · arrow 0.3.3 · jdbc 0.3.3.

Companion PR in softclient4es-web sweeps the same numbers on the public site.

Verified before writing, not after

Every version written here returned 200 on JFrog first — jdbc + adbc drivers for es6/7/8/9 at 0.3.3, softclient4es8-java-client at 0.23.0, arrow-extensions 0.3.3, community-extensions 0.3.2. That check exists because an earlier sweep shipped a doc that installed a version which never shipped, and the failure is silent.

Per-occurrence triage, not a blanket replace

Deliberately NOT bumped — each of these looks identical to a regex and means something different:

reference	why it stays
joins.md "since arrow-extensions 0.2.5 (REPL bundle 0.20.4, driver 0.2.5)"	a since line — when the behaviour appeared
joins.md "Since engine 0.23.0 with arrow-extensions 0.3.3"	already the correct pair for this train
the subquery / CTE limitation statements	still true — see the honesty note below

Found while sweeping

• README was stale by TWO releases. Its JDBC driver table and all three coordinate blocks still said 0.3.1 while documentation/client/jdbc.md said 0.3.2 — the 0.3.2 sweep had missed README entirely.
• community-extensions was held at 0.3.1 deliberately last train (the extensions repo had not moved). It has now, so it bumps to 0.3.2 here — a deliberate non-bump becoming a bump.

install.cmd — the release ritual, which is functional, not documentation

PS1_REF v0.22.0 → v0.23.0 and PS1_SHA256 recomputed from the real tag (9a7c772c…). A bumped ref with a stale hash is a hard failure for every install.cmd-only user, so the two always move together.

Method controlled first: the hash pinned before this change matches v0.22.0's install.ps1 byte for byte, which is what makes the new value trustworthy.

🔴 Patched as bytes. .gitattributes marks *.cmd -text, so a text-mode rewrite silently normalises CRLF and turns a 3-line edit into 330. The diff here is 3 lines and the file is still CRLF.

repl.md's version listing was re-captured, not retyped

It is quoted product output, so it was regenerated by actually running ./install.sh --list-versions --es-version 8. It now shows 7 versions ending at 0.23.0.

That run also proves the softclient4es8-cli-all_2.13 bundle published at 0.23.0 — worth stating, because a missing -all bundle falls back to the plain artifact, which carries no arrow extensions and therefore no cross-index JOIN. The docs would demonstrate the path that loses the headline feature, silently.

⚠️ Honesty note on story 22.1

Nothing here claims derived tables or subqueries execute. 22.1 makes FROM (SELECT …) AS d and JOIN (SELECT …) AS d ON … parse and validate, then be refused with HTTP 400 wherever the relational engine is absent. Execution is story 22.4. The existing "no arbitrary subqueries" / "subqueries, CTEs … not supported" statements remain correct and are untouched.