Port the SDKs to Effect 4.0.3 and sui-effect 0.3.0

[tamashi095]

Summary

Ports the Effect-based SDKs from effect@4.0.0-rc.112 + @unconfirmed/sui-effect@0.2.x to stable Effect 4.0.x (peer floor 4.0.3, developed and tested on 4.0.4) + @unconfirmed/sui-effect@0.3.0. Prepares the release only: no tags are pushed and nothing is published here.

Package	Old	New
@misofm/streaming	0.3.0	0.4.0
@misofm/musicos	0.7.1	0.8.0
@misofm/partyos	0.7.1	0.8.0
@misofm/platform	0.45.0	0.46.0
@misofm/transcoding	0.4.0	0.5.0
@misofm/effect (deprecated)	0.2.0	untouched, no release

What changed

• Peers: effect >=4.0.3 <4.1 (transcoding also @effect/platform-node), @unconfirmed/sui-effect >=0.3.0 <0.4.0, @mysten/sui ^2.29.0 (was exactly 2.29.0 in partyos/platform; musicos keeps ^2.28). Dev pins: effect and @effect/platform-node 4.0.4 (a 4.0.3 pin is an invalid peer tree: platform-node-shared ^4.0.3 resolves to 4.0.4, which peers effect ^4.0.4), sui-effect 0.3.0, @mysten/sui 2.29.0. The isolated consumers install with --strict-peer-deps and run npm ls.
• @mysten/bcs is now a peer (^2.1.1) in partyos and platform instead of a dependency, so one copy is installed.
• Source: the only compile break was Config.nonEmptyString -> Config.NonEmptyString in Partyos.layerConfig, Musicos.layerConfig and Miso.layerConfig. The published .d.ts files of every package are identical to the previous release (diffed), so there is no public API change.
• partyos: new tests/golden.test.ts builds both party recipes offline and asserts byte-identical TransactionData and digests for four golden vectors generated by 0.7.1 on rc.112.
• @misofm/effect stays frozen on rc.112 (superseded by sui-effect, nothing depends on it). The consumer fixture and check-consumer skip it because its exact rc peer cannot share an install with the ported packages.
• Each ported package gets a CHANGELOG.md (shipped in the tarball) with a consumer migration note and the behaviour changes from Effect 4.0.x / sui-effect 0.3.0.

Behaviour audit

No partition/separate, Effect.timeout, Effect.retry, Schema.brand or Encoding use in these packages. TimeoutError now has a message and Effect.retry skips mixed typed+defect causes (both reach callers only through sui-effect). Interruption semantics changed in 4.0.x for the transcoder (checked by its cancellation test). A differential run over every exported error class against the rc.112 build shows identical message, JSON, Schema encode/decode output. Details are in the per-package changelogs.

Verification

bun install --frozen-lockfile, bun run typecheck, bun run test (streaming 27, effect 23, partyos 41, musicos 93, transcoding 52, platform 386, all pass), bun run build, npm pack --dry-run per package, bun run test:consumer (npm install of the packed tarballs into the isolated fixture: one effect, one @mysten/sui, one @mysten/bcs), the transcoding isolated Node 22 and Bun consumers, and the ffmpeg integration test with a local FFmpeg 8.1.2 (including MISO_PINNED_FFMPEG=1). The workspace tests also pass with @mysten/sui 2.35.0. codegen:check was not run: it needs the sui binary and the sibling Move checkouts, no generated file changed, and CI does not run it.

CI

New sui-matrix job runs partyos, musicos and platform typecheck and tests against @mysten/sui 2.29.0 and 2.30.0.

Release (after merge)

Tag the merge commit on main, in dependency order: streaming-v0.4.0, then musicos-v0.8.0 and partyos-v0.8.0, then platform-v0.46.0 and transcoding-v0.5.0.

Risks for consumers

See the changelogs. The main ones: rc consumers (misofm/api, misofm/app) must move effect and every @effect/* package to 4.0.x and sui-effect to 0.3.0 together; @mysten/bcs must now be installed by the consumer; pin effect and every @effect/* package to one 4.0.x (the changelogs show 4.0.4); the peer range alone can resolve an invalid tree.

🤖 Generated with Claude Code