MariaDB: share writable-directory setup across VFS images

[brandonpayton]

What changed

#907 already put the shared MariaDB writable-directory helper and its serialized-image ownership regression on main.

This PR keeps the small part that was still unique:

• make the LAMP image use that shared helper;
• declare the helper as a LAMP cache input; and
• remove the unused browser-only MariaDB directory helper and export.

What I removed

I dropped the duplicate ownership implementation, the already-landed helper and test, and the stale MariaDB/LAMP revision changes from the old branch. The old head was ebb2a09f3b34e0d00b68791c207caa52d3192018.

ABI and package artifacts

There is no ABI change; the ABI 39 snapshot and generated bindings remain unchanged.

The LAMP cache key now includes the helper file, but the helper refactor does not change the directory bytes: the old inline setup and shared helper produced byte-identical raw VFS images (2f4e0cb3... for both). Full LAMP rebuilds had the same 8,710 entries, image metadata, file types, modes, owners, groups, and sizes. Rebuild-to-rebuild content drift was limited to generated MariaDB/WordPress database files and one OPcache file. Package revision stays at 8.

Validation run

All commands ran through scripts/dev-shell.sh.

• host TypeScript check;
• MariaDB image helper and VFS ownership tests: 36 passed;
• package build-input check: 14 tools across 7 consumers passed;
• wasm32 and wasm64 musl builds;
• ABI snapshot/version check;
• browser asset check: 92 imports resolved;
• source-built LAMP image on both old and new setup paths, with the byte/metadata comparison above;
• Node-hosted MariaDB initialization, WordPress installation, and OPcache prewarm during both LAMP builds;
• ./run.sh browser reached the Vite server; the WordPress MariaDB page was inspected and showed the preinstalled site running;
• focused Chromium merge-gate regression: WordPress MariaDB booted and logged into wp-admin (1 passed).

I also ran the older slow WordPress login test. It reached the login form but timed out because its direct submit click is covered by the guide popover; the maintained merge-gate test uses the guide login action and passed. That failure is unrelated to this diff.

Validation not run

Firefox, WebKit, the full slow browser suite, external libc/POSIX/Sortix suites, and performance benchmarks were not run. This is not a performance change and makes no performance claim.

[github-actions]

Phase B-1 matrix build status — pr-896-staging

ABI v39. 1 built, 0 failed, 1 total.

Package	Arch	Status	Sha
lamp	wasm32	built	c2b144d6

_{Auto-generated; replaced on each push. Raw data in the publish-status workflow artifact.}

[brandonpayton]

Before the post-#907 rewrite, this PR pointed at ebb2a09f3b34e0d00b68791c207caa52d3192018. I verified that SHA again and found no open PR based on this branch. I am replacing the stale ownership/revision work with the small remaining helper-sharing diff directly on current main (92d5940f7e0107514ea12ab813d395257678377e) using an exact force-with-lease.

[brandonpayton]

Closing this focused source PR because #934 now carries LAMP’s use of the shared MariaDB helper, its cache input, and the dead browser-helper removal at exact green head 260d75d5adc31b327f5591229dbfbb287d395fd6. The helper implementation and serialized-image coverage already on main through #907 were deliberately excluded, and no unnecessary revision bump was added. The full staging and browser gates passed. This is not a claim that this branch was merged verbatim.