Migration notes for akm v0.7.4

This is a patch release that fixes the publish process. No functional changes from v0.7.3.

0.7.4 includes the same features as 0.7.3:

• akm index --enrich opt-in for LLM passes — index-time enrichment work such as metadata enhancement, memory inference, and graph extraction now runs only when explicitly requested with --enrich. Default indexing is faster and no longer surprises operators with LLM-backed work during normal maintenance runs.
• Config backup snapshots before writes — config writes now create AKM cache backups so setup/config flows have a recovery path if a config is overwritten or corrupted during development or testing.
• Setup wizard UX refresh — akm setup now better reflects the real configured state: source prompts are ordered more sensibly, configured and preserved stash information is surfaced, agent defaults can be selected explicitly (including disabled), and post-setup indexing does not implicitly enable enrichment.
• CI workflows updated for current GitHub Actions runtimes — CI, release, and publishing workflows now use current action majors (checkout@v5, cache@v5, setup-node@v5, upload-artifact@v5, download-artifact@v6) to stay off deprecated Node 20 action runtimes.
• Embedding-dimension drift on read-only DB opens — read/telemetry paths no longer mutate the live index schema with the default embedding dimension. akm info, search/show parity paths, and related readers now preserve the configured embedding shape instead of downgrading vector tables.
• Incremental index churn across multiple source layouts — incremental indexing is now significantly more stable for filename-less legacy metadata, wiki-root sources, repo-root git stash layouts, non-indexed companion files, and cross-source dedupe cases.
• Git source indexing for repo-root stashes — git-backed sources no longer assume a <repo>/content subtree; repo-root stash layouts are indexed correctly and cached mirrors are treated as fresh instead of being needlessly refreshed.
• show metadata no longer depends on .stash.json — command and skill summary/show metadata now comes from file-local frontmatter and renderer parsing rather than the deprecated disk fallback sidecar.
• .stash.json no longer drives incremental stale detection — editing .stash.json alone no longer forces directories to rescan during incremental indexing.
• One-shot URL ingest for akm import and akm wiki stash — both commands now accept a single HTTP/HTTPS URL in addition to file paths and stdin. akm import <url> fetches the exact page, converts it to markdown, and writes it into knowledge/ using a URL-path-derived default name. akm wiki stash <wiki> <url> fetches the exact page, converts it to markdown, and writes it into wikis/<wiki>/raw/. Neither command registers a persistent website source or crawls linked pages.

For stash authors on the 0.7.x pre-release line: .stash.json remains supported for compatibility in this release, but it is deprecated and will be removed in v0.8.0. That timeline is intentional: during this aggressive pre-release phase-out window, compatibility shims do not stay around until 1.0 unless they still earn their cost. Prefer frontmatter for markdown assets and structured code comments for scripts, and migrate any remaining .stash.json metadata before taking the 0.8 upgrade.