Rewrite to Typescript

[rzcoder]

Summary

Full rewrite of data-struct in TypeScript with a new codec-token API,
two-pass encoder, dual ESM/CJS build, vitest + benchmarks + CI, plus a
follow-up security pass and a perf pass on the field-iteration hot path.
Wire format is byte-identical with 0.0.x for the corresponding new
codecs; legacy DataTypes / DataReader / DataWriter exports are
removed (clean break, ships as 0.1.0).

Changes

API

• Typed codec tokens: t.bool, t.i8/u8/...i32/u32, t.f32/f64,
  t.i64/u64, t.string, t.shortBytes, t.bytes, with t.le.*
  little-endian variants.
• struct(schema) factory that compiles a schema once and reuses it
  across encode / decode / sizeOf.
• Functional encode(value, schema) and decode(buf, schema).
• Infer<S> derives the TypeScript shape of decoded values from a
  schema (including top-level array forms — pinned by tests).
• DataStructError with codes (VALUE_OUT_OF_RANGE, STRING_TOO_LONG,
  BYTES_TOO_LONG, ARRAY_TOO_LONG, BUFFER_UNDERFLOW,
  SCHEMA_MISMATCH, INVALID_SCHEMA), field path (e.g.
  $.skills[1].description), and decode offset.

Build & packaging

• Dual ESM + CJS via tsup with separate .d.ts / .d.cts. Minified
  output: dist/index.mjs is ~7.5 KB (down from ~15 KB).
• "type": "module", engines.node >= 20.9, sideEffects: false.

Performance

• Two-pass encoder allocates the output buffer exactly once (no
  Buffer.concat).
• Lazy error-path construction via try/catch + rethrowWithPrefix
  instead of allocating ${path}.${key} strings per field. Measured on
  Node 22: list of list encode -59% / decode -45%, nested encode
  -33%, hero encode/decode -17% each.

Security hardening

• Decoded structs use Object.create(null) — neutralises prototype
  pollution from a __proto__ key in an untrusted schema.
• String decoder is TextDecoder({ fatal: true }) — invalid UTF-8 now
  raises SCHEMA_MISMATCH instead of silently substituting U+FFFD.
• struct.encode asserts measure() bytes == write() bytes, so a
  buggy codec can't leak uninitialised memory from Buffer.allocUnsafe.
• CI runs npm audit --audit-level=high --omit=dev.

Tooling & CI

• vitest suite: wire-format goldens, roundtrip, error paths, v8
  coverage.
• tinybench benchmark suite + on-demand benchmark.yml workflow.
• biome for lint + format.
• CI matrix: Node 20/22/24 on Linux + Node 22 on macOS/Windows.
• Tag-triggered release workflow with npm provenance.
• Dependabot, PR template, CODEOWNERS, .editorconfig, .nvmrc.

Examples & docs

• examples/01..06-*.ts covering basics, the hero struct, functional
  form, errors, little-endian, and sizeOf.
• README rewritten; CHANGELOG.md added.

Testing

• npm test passes (vitest: wire-format goldens, roundtrip, errors)
• npm run lint passes (biome)
• npm run typecheck passes
• npm run bench — perf numbers reported above

Notes

Breaking changes. This ships as 0.1.0. Legacy DataTypes,
DataReader, and DataWriter exports are gone; consumers move to
struct(...) / encode / decode / t.*. The README has a migration
table. Minimum Node is now >=20.9.

Wire format. Byte-identical with 0.0.x for the corresponding new
codecs — the previous test suite's golden buffers are preserved in
test/wire-format.test.ts to lock this in.

Removed dev deps. grunt, grunt-simple-mocha, jit-grunt,
grunt-contrib-jshint, chai, benchmark.