0.4.0 release prep: ship FORMAT.md, presence docs, error hygiene

[deepfates]

Pre-publish fixes for the 0.4.0 npm release, from a cold-read of the package as a stranger would install it. Prep only — do not merge or publish without the owner's call.

1. Ship the spec in the tarball

files was ["bin", "dist"], so the README's links to FORMAT.md and pacts/*.md were dead from node_modules. Added FORMAT.md and pacts to files.

npm pack --dry-run (dist/* elided — 136 files, unchanged):

npm notice package: @deepfates/lync@0.4.0
npm notice 22.2kB FORMAT.md
npm notice 1.1kB LICENSE
npm notice 20.6kB README.md
npm notice 49B bin/lync.js
npm notice 5.4kB package.json
npm notice 1.2kB pacts/authorship.md
npm notice 6.7kB pacts/export.md
npm notice 9.5kB pacts/import.md
npm notice 690B pacts/ordering.md
npm notice package size: 116.5 kB
npm notice unpacked size: 501.6 kB
npm notice total files: 145

2. Document ./presence-awareness

The export existed with no README surface. New "Presence: who is here right now" section with a runnable example (two awareness instances wired to each other; the SyncedStore wiring is stated in prose), plus a line in Subpath exports. The example carries no fragment directive, so check-readme-examples executes it in pnpm verify and CI — it runs as example 6.

3. Non-empty socket error reason

lync sync against a dead relay printed socket error from ws://...: with nothing after the colon — Node's built-in WebSocket fires an ErrorEvent whose message can be the empty string. New socketErrorReason in src/cli/sync.ts falls back to the underlying error's message, then its code (e.g. ECONNREFUSED), then its name, and never returns empty. Covered by a dead-relay integration test (relay started, port noted, relay killed, sync attempted) asserting the reason after the colon is non-empty, plus unit tests over every event shape.

4. README status line

Top of the README now states: format is a v0 draft (matching FORMAT.md's own status line), envelope v:1 is stable.

5. "Why not Yjs / Automerge?"

Short honest section grounded in FORMAT.md's actual argument: events are immutable so there is no concurrent-mutation problem; merge is set union by id with no merge logic; same-id-different-bytes is surfaced, not resolved; CRDTs solve a different problem and are the right choice when state mutates. Notes that presence uses LWW clocks, not a CRDT.

Verification

• pnpm test: 21 test files, 125 tests, all passed (includes the 2 new socket-error tests).
• pnpm verify (path guard + tests + build + typecheck + README examples): passed. Examples: 9 executed (including the new presence example), 4 skipped as declared fragments.
• npm pack --dry-run: file list above; FORMAT.md and all four pacts included.