fix(compose): mount named volumes so data survives --force-recreate

[itxtoledo]

Problem

Named volumes declared in a docker-compose.yaml are silently skipped by mocker compose up. The volume store directory is created (you can see it under ~/.mocker/volumes/<project>-<name>/_data), but it is never bind-mounted into the container.

The consequence is severe for stateful services:

• Writes go to the container's ephemeral layer, not to the volume.
• Any container recreation — compose up --force-recreate, a changed config hash, compose down && compose up — discards the data.
• MySQL, Postgres, Redis, etc. come back empty with no warning at all.

I hit this in production dev: a MySQL stack lost its entire database on --force-recreate, and every named volume under ~/.mocker/volumes/ was 0 bytes — proof they were never written to.

Root cause

ComposeOrchestrator.resolveVolumeMounts (ComposeOrchestrator.swift:564) intentionally drops named volumes:

// named volumes (bare names without path separators) are skipped — Apple's
// virtiofs doesn't support chown from within containers, which breaks images
// like postgres that chown their data directory on init.

The skip was added in c0934d8 as a revert of a87cf87 (which had the right idea). The virtiofs/chown concern is real but applies equally to regular bind mounts, which compose already supports and which demonstrably work — including for postgres-style images. Silently losing data on every recreation is the worse failure mode, and the volume store directory is owned by the host user (exactly what mocker run -v name:/path already relies on).

Fix

Named volumes are now resolved to their backing directory and bind-mounted, exactly like Docker does internally:

• name:/container/path → -v <volumesPath>/<runtimeName>/_data:/container/path
• runtimeName applies the project prefix unless the volume declares an explicit name: or is external: (reusing the existing ComposeVolume.runtimeName(projectName:) logic, so compose down --volumes removal stays in sync).

Changes

File	Change
Sources/MockerKit/Compose/ComposeOrchestrator.swift	resolveVolumeMounts now takes projectName, declaredVolumes and volumesPath; declared named volumes resolve to <volumesPath>/<runtimeName>/_data. Undeclared bare names are still dropped (previous behaviour).
Sources/MockerKit/Volume/VolumeManager.swift	Expose mountpointPath (nonisolated) so compose can resolve volume backing dirs.
Tests/MockerKitTests/ComposeOrchestratorTests.swift	Replaced the old "named volume skipped" expectation with coverage for declared, custom-named, external, undeclared, and mixed specs.
CHANGELOG.md	Unreleased → Bug Fixes entry.

Verification

• Unit: swift test — 448 tests in 35 suites, all green.
• Integration (alpine with a named volume):

  services:
    app:
      image: alpine:3.20
      command: ["sh", "-c", "echo hello > /data/hello.txt && sleep 5 && cat /data/hello.txt"]
      volumes:
        - mydata:/data
  volumes:
    mydata:

  1. mocker compose up → hello.txt appears in ~/.mocker/volumes/<project>-mydata/_data/ (previously the volume stayed empty).
  2. mocker compose up --force-recreate → hello.txt survives (previously the data was lost).
  3. mocker compose down --volumes → volume removed, as expected.

Notes / trade-offs

• Images that chown their data dir on init (e.g. official postgres) may need the host directory to be writable by the container user — the same constraint regular bind mounts already impose today. If virtiofs chown support is added later, no further changes are needed here.
• Behavior is now consistent with Docker Compose: named volumes persist across recreation and are removed only by down --volumes.