v1.6.1: gjson-style fast-skip optimization + benchmarks vs gjson/sonic

Performance:
- gjson >'\' fast-skip in stringEndConfig, blockEndConfig, searchKeysConfig
- Medium: -19.4%, Small: -11.3%, Large: -3.2% (zero allocations maintained)

Benchmarks:
- Added gjson (tidwall/gjson) and sonic (bytedance/sonic)
- jsonparser is now the fastest across ALL payload sizes
- README updated with full comparison

v1.6.0: Append function + all KI fixes (zero open known issues)

New API:
- Append(data, value, keys...) — clean array-append without knowing
  array length. Works on top-level and nested arrays. Auto-creates
  missing paths as single-element arrays.

Bug fixes (all known issues resolved):
- KI-2: ParseInt('-') now returns MalformedValueError (was 0, nil)
- KI-3: disposition updated to 'fixed' (already fixed via auto-coerce)
- KI-4: Set on top-level array index now appends (was KeyPathNotFoundError)

Proof: 123 requirements, 0 errors, 0 warnings, 0 open known issues.
All 4 KIs now status: fixed.

Contributed by codex (gpt-5-codex) via codex exec.

v1.5.1: 6.1x large-payload speedup + fresh benchmarks

Performance:
- Fix stringEnd unbounded backslash scan (5.8x speedup)
- SWAR string scan (additional 8%)
- Total: 128us -> 21us on large payload

Benchmarks:
- Updated ALL comparison libraries to latest versions
- Fixed ffjson measurement bug (encoding/json now measured correctly)
- README benchmark tables refreshed with real data
- Methodology documented (M4 Max, Go 1.26.3, median of 5 runs)

README headline: 'up to 6x faster' (was '10x' — old number measured
ffjson not encoding/json)

CHANGELOG: v1.5.1 entry with performance table.

docs: complete CHANGELOG with v1.3.0–v1.5.0 entries, all mentioning R…

…eqProof coverage

v1.4.0: ArrayEachErr, Escape/SetString, wildcards, JSONPath, GetArray…

…Len, GetUint64, DeleteFound

New APIs (all backward-compatible, additive only):

Iteration with error/break control:
- ArrayEachErr: callback returns error to stop iteration early (io.EOF = graceful stop)
- EachKeyErr: same pattern for EachKey
  Resolves: #53, #129, #176, #230, #255, #262

Safe string handling:
- Escape: RFC 8259 string escaping (inverse of Unescape)
- SetString: Set with auto-quoted string value
  Resolves: #144, #158, #218, #270

Container accessors:
- GetArrayLen: count array elements without callback
- GetObjectLen: count object key-value pairs without callback
- GetUint64: uint64 variant of GetInt
  Resolves: #175, #261, #271

Delete found signal:
- DeleteFound: returns (result, found bool)
  Resolves: #229

Wildcard paths:
- EachKeyWildcard, ArrayEachWildcard, SetWildcard: [*] path component
  Resolves: #112

JSONPath compiled paths:
- ParsePath: $.a.b[0] → []string path
- CompilePath + CompiledPath: pre-compile + Get/Set/Delete methods
  Resolves: #234, #251

Fixes:
- EachKey no longer panics with >64 key components (#56)
- Set pre-allocates output buffer reducing allocations 6→1 (#107)

Proof coverage:
- 3 new SYS-REQs (112 container length, 113 wildcard, 114 compiled paths)
- 121 total requirements, 0 errors, 0 warnings
- 384 MC/DC witness rows, 0 uncovered
- LocalChange CHG-260728-RKTS records the feature release
- All new functions traced via source-native annotations

Contributed by codex (gpt-5-codex) via codex exec.

v1.3.1: fix Set aliasing (#209/#141), EachKey array-index (#232), ben…

…chmark ffjson (#126)

Bug fixes:
- Set/Delete no longer mutate the caller's input buffer backing array
  (append into spare capacity). Fixes #209, #141.
- EachKey now descends into terminal array-index paths consistently with
  Get. Fixes #232.
- Benchmark payload types no longer have ffjson-generated methods; the
  encoding/json benchmark now measures the real stdlib. Fixes #126.

Proof strengthening (closes the gaps that let these bugs escape):
- New obligation: no_input_mutation (Set/Delete must not modify input
  beyond the returned slice) — catalog overlay + proof.yaml declaration
- New obligation: api_consistency (EachKey path resolution == Get path
  resolution) — catalog overlay + proof.yaml declaration
- assertInputUnchanged gate in property/oracle tests: snapshots input
  bytes + backing-array capacity before every Set/Delete, verifies
  unchanged after
- TestApiConsistencyEachKeyMatchesGet gate: random JSON + paths,
  asserts EachKey result == Get result (the differential that would
  have caught #232)
- Benchmark honesty lint: verifies no benchmark type implements
  json.Marshaler/Unmarshaler (would have caught #126)

Contributed by codex (gpt-5-codex) via codex exec.

docs: add proof coverage to README + CHANGELOG for v1.3.0

README: prominent L3 assurance badge + proof-coverage section (118 reqs,
0/0 audit, 100% MC/DC, 250k fuzzer execs/sec) with link to reqproof.com.
Positions jsonparser as the ReqProof reference case study.

CHANGELOG.md: full v1.3.0 entry featuring the formal-verification work,
7 bug fixes, 2 performance improvements, contributor credits, and links
to the json-fuzz package and root-cause analysis.

Merge pull request #269 from buger/tinygo

Add support for tinygo

Merge pull request #276 from dbarrosop/master

fix: prevent panic on negative slice index in Delete with malformed JSON (GO-2026-4514)

Merge pull request #221 from d-hat/CVE-2020-35381

Attempt to fix #219 and introduce a test.  The only error that can easily be returned in this case is `KeyPathNotFoundError`, which is reasonable if you squint (a malformed key can not be found).

Note I'm far from fluent in golang so this should be reviewed with some care 😄