docs: document the CI-artifact route for the release archive - #323

Adds a subsection to docs/RELEASE.md § Release Export covering what to do when :app:exportReleaseToBuildDrive cannot run locally.

The local export stays the default and the section says so. But it is the step most likely to be the one a given machine cannot do: minifyReleaseWithR8 is the heaviest task in the build, and short of memory it does not fail cleanly — it kills the Gradle daemon, and the message names neither memory nor R8. v2.32.6 hit this at both -Xmx2048m and -Xmx1536m with ~1.7 GB free, and the archive came from the tag job's artifacts instead.

That substitution is sound — release runs the same task on the same tagged commit with the same signing key — but it was decided ad hoc during a release, which is the wrong time to be deciding whether something counts. The section covers the symptom and how to recognise it, the download and naming, assembling the notes file from the fastlane changelogs, verifying with verify-release-export.ps1, and a jarsigner check for the release certificate (the one property the local build gives for free and this route does not).

It also names the two things this route changes:

• Ordering inverts. The gate exists to run before the tag; artifacts only exist after it, so the check that was meant to catch a forgotten export becomes the step that can be forgotten.
• It expires. Workflow artifacts last 30 days, and a rebuilt R8 mapping will not deobfuscate traces from the shipped binary.

Commands were verified against the v2.32.6 archive: the run lookup resolves to 31891151897, and the notes loop reproduces markleaf-v2.32.6-vc128-release-notes.txt exactly (Compare-Object → no differences).

No change to the PowerShell quoting guidance — -Pmarkleaf.requireReleaseSigning=true is already documented as needing quotes in § Local Signed Release Build, which is what I should have followed.

🤖 Generated with Claude Code