docs: bring user-facing versions up to the 0.20.2 release train

[fupelaqu]

Brings every user-facing version reference in the core docs up to what the 0.20.2 release train actually published.

Versions

Artifact	Was	Now
engine / softclient4es8-java-client	0.20.0	0.20.2
softclient4es-community-extensions	0.2.0	0.2.2
softclient4es-arrow-extensions	(absent)	0.2.3
softclient4es{6,7,8,9}-jdbc-driver	0.2.1	0.2.3
softclient4es{6,7,8,9}-adbc-driver	0.2.1	0.2.3
softclient4es{6,7,8,9}-arrow-flight-sql	0.2.1	0.2.3
softclient4es-jdbc-driver_2.13 (%% form)	0.2.1	0.2.3

Every pair was checked against the JFrog listing API before editing. Note the deliberate asymmetry: community-extensions stops at 0.2.2 — 0.2.3 does not exist for it — while arrow-extensions is at 0.2.3. Review independently confirmed both, and that arrow-extensions 0.2.3's POM does not depend on community-extensions, so the mixed pairing is safe.

Beyond the mechanical bump

repl.md contradicted itself three ways. The --list-versions sample advertised "0.20.2 / 0.20.3" (0.20.3 never shipped), and the Installation Examples thirty lines below then installed 0.16.0. So one page showed three different ideas of what exists. The sample now lists the real bundle line (0.20.1 / 0.20.2) and the examples install 0.20.1.

This is not cosmetic. Per install.sh, a version with no matching -all bundle logs No -all bundle for version … falling back to the plain artifact, and per repl.md's own text the plain artifact has no arrow extensions and therefore no cross-index JOIN. The docs were handing users a copy-paste command that silently loses the headline feature.

README's sbt block never mentioned arrow-extensions, only community-extensions — while repl.md calls arrow-extensions required for cross-index JOIN. Added, with the Java 11+ note.

install.sh: the --help example pinned 0.20.0 (same no-bundle fallback trap) and a comment still referenced the phantom 0.20.3.

Deliberately unchanged

Historical statements: documentation/sql/* "since v0.14.0" feature markers and CONTRIBUTING's release-process example describe the past, not what to install. Rewriting them would make them false.

Verification

Review swept both this repo and the web repo over all tracked files and found no remaining stale reference; confirmed every bumped pair resolves on JFrog; and checked MD↔MDX parity (versions, artifact names, Java floor, install commands all agree). The one divergence it found — arrow-extensions described as "optional" on the web while this PR calls it required — is fixed in the companion web PR.

The install.ps1 stale example the review also flagged is fixed in #180 instead, which owns that file, to avoid a merge conflict.

Companion

softclient4es-web docs/versions-0.20.2 — same bump for the public site, including the four download buttons and their data-version analytics attributes.

[fupelaqu]

Companion web PR: SOFTNETWORK-APP/softclient4es-web#29 (same bump for the public site, incl. the four download buttons and their data-version analytics attributes). Merge them together so the dual-maintained MD/MDX copies don't drift.

[fupelaqu]

Rebased onto main (now that #180 is merged) and picked up one extra commit.

That second commit fixes install.ps1's --help example, which pinned 0.20.0 — a version with no -all bundle, so the example quietly demonstrated the fallback to the plain artifact with no cross-index JOIN. It was originally pushed to fix/179 but landed after #180 was merged, so it never reached main; it rides here instead.

Companion web PR SOFTNETWORK-APP/softclient4es-web#29 is already merged, so merging this closes the MD/MDX gap.