[0.1.7] Add GUI QA driver

[unurgunite]

Description

Adds a host-side GUI test driver (tools/gui-qa/) that drives the plugin inside RubyMine running in a Tart VM and verifies behavior end to end: palette/menu actions, notification balloons, editor content, and idea.log oracles. Covers sections 3D (RBS detection), 3E (Update Types), 3F (workspace batch), 3G (Doctor), and 3H (MCP/settings/folding/i18n) — driver and docs only, no production code.

How it works:

• lib.sh — SSH transport (VM IP resolved per run, proxy vars stripped), screenshots over scp, Vision OCR assertions, Search Everywhere handling (opened via menu clicks — keyboard chords get intercepted by the macOS man-page service and steal focus), frontmost-app gates before every keystroke batch.
• stand.sh / ws.sh / menu.sh / anno.sh / ut.sh / settings.sh — project-stand bootstrap, workspace-run helper (log-gated: waits for a fresh collectRubyFiles line before accepting a balloon, so stale notifications can't false-pass), context-menu firing, annotator log oracles, Update Types helpers, Settings helpers (triple-click label + Space for checkboxes — the square has no OCR row; clipboard reads via in-balloon drag, not Cmd+A).
• cases/3*.sh — one script per checklist item, each ending in ok / not ok with md5/log/balloon oracles.
• merge.py — merges case outcomes into qa-results-auto.json by section order.

Results (recorded in qa-results-auto.json in the gem repo, 192 ok / 7 fail / 3 skip of 202): 3D 7/7, 3E 8/8, 3F 5/6, 3G 5/5, 3H 6/8. Remaining items are documented scope cuts or drifts in the case files: no live MCP port in the VM (3H.2, registration + code shape asserted instead), 33- and 207-file workspace checks finish too fast for any live cancel (3F.4, cancel plumbing stays unit-tested), warn-toggle OFF still flags syntax-broken YARD (3H.5 — gem always emits invalid_type, toggle not wired into CHECK), plus the pre-existing 3C gem drifts.

Also includes RUBYMINE-PILOT.md with the pilot findings the driver is built on.

Related Issue

N/A — test infrastructure, no user-facing change.

Checklist

• Driver executed live against RubyMine 2026.2 (3D 7/7, 3E 8/8, 3F 5/6, 3G 5/5, 3H 6/8; outcomes merged into qa-results-auto.json)
• Tests pass (./gradlew test) — N/A, no production code touched
• Formatting clean (./gradlew spotlessCheck) — N/A, shell/markdown only
• Static analysis clean (./gradlew detekt) — N/A
• Plugin verified (./gradlew verifyPlugin) — N/A
• CHANGELOG updated — N/A, no user-facing change

Screenshots (if applicable)

N/A — assertions are OCR text plus logs; shots stay in /tmp/gui-qa-rm/ by design and are never committed.