chore(release): prepare LeanTypeDual 0.3.0

[AsafMah]

Version bump and release paperwork for 0.3.0. Everything mechanical is done; the tag and publish are deliberately left for after real use.

Version

0.3.0 / versionCode 4300, per the fork's offset formula 4000 + major*1000 + minor*100 + patch*10. Monotonic above 0.2.0's 4200, so upgrades keep working.

Minor rather than patch: a 200-commit upstream merge with real new features, not a bugfix roll-up.

Contents

• LeanBitLab/LeanType v4.1.2 (Merge upstream LeanBitLab/LeanType v4.1.2 #137) — clipboard edit mode, voice recognition language selector (auto-detect + 99 languages), personal-dictionary learning threshold, physical-keyboard suggestion shortcuts, localized Compose settings, text-expander regex fixes.
• Two fork regressions the merge would otherwise have shipped (Merge upstream LeanBitLab/LeanType v4.1.2 #137) — whole-word backspace leaving partial words behind, and custom layouts no longer restoring after symbol mode. Caught only by this repo's own tests, not by the compiler.
• Accelerated-backspace emoji fix (fix(input): correct the accelerated backspace emoji measurement #134) — the second deletion measured the code point from before the first deletion and could split an emoji. Also present upstream; reported as LeanBitLab/LeanType#423.
• Test-result gate (chore(tools): gate test-result analysis instead of trusting it #139).

Adds fastlane/metadata/android/en-US/changelogs/4300.txt and docs/releasenote/release_notes_v0.3.0.md, and renames [Unreleased] → [0.3.0].

Signed artifacts built and verified

The release workflow was run against this branch (run 32348509205). Because the publish step is gated on startsWith(github.ref, 'refs/tags/'), this produced signed APKs and no GitHub Release — v0.2.0 is still latest.

All four verified after download:

Flavor	applicationId	version	minSdk	INTERNET	signer
standard	com.asafmah.leantypedual	0.3.0/4300	23	yes	✅ c032eafc…
standardfull	com.asafmah.leantypedual	0.3.0/4300	23	yes	✅
offline	…leantypedual.offline	0.3.0/4300	26	no	✅
offlinelite	…leantypedual.offlinelite	0.3.0/4300	21	no	✅

The signed standardfull build is installed on the maintainer's device as com.asafmah.leantypedual, alongside the debug and experimental builds. That closes the gap a debug-only smoke test leaves open: release adds R8 minification and resource shrinking, the classic source of "worked in debug, broke in release".

Note: a fresh install of the release package starts with no user-supplied gesture library in its private files/, so glide typing needs the library loaded before it can be judged.

Other verification

• fork invariants: applicationId, INTERNET only in standard/standardfull
• compileOfflineRunTestsKotlin passes
• :app:testOfflineRunTestsUnitTest: 328 tests, 4 failed — the documented Windows-only ParserTest baseline, confirmed by tools/check_test_results.py
• dev CI green after every merge that feeds this

⚠️ Still not done, on purpose

No tag, no publish. The artifacts exist and are verified; what's missing is somebody actually typing on it. Device smoke is a gate HANDOFF.md lists and that has never been executed for a release, and this session demonstrated the cost of skipping it — a backspace fix shipped on plausible code analysis turned out to address a bug that does not reproduce on the device at all.

Once the build has had real use:

# merge this PR, promote dev -> main via HANDOFF section 9 (read-tree recipe)
git tag v0.3.0 && git push origin v0.3.0
gh workflow run release.yml --repo AsafMah/LeanType --ref v0.3.0   # now creates the draft
gh release edit v0.3.0 --repo AsafMah/LeanType --draft=false --latest \
  --title "LeanTypeDual 0.3.0" --notes-file docs/releasenote/release_notes_v0.3.0.md

Not included

#141 (experimental two-thumb) is intentionally out. It's opt-in and defaults to current behaviour, but it exists to be A/B'd against a working daily driver, not folded in beforehand.