fix: append scalars when Set targets array index beyond length

[Solaris-star]

Summary

Set with successive array indexes on a newly created path overwrote the whole array instead of appending:

Set({}, 1, "test", "[0]") // {"test":[1]}
Set(..., 2, "test", "[1]") // was {"test":[2]}, want {"test":[1,2]}

Root cause: when extending an existing non-empty array, only arrays whose next token was { were treated as appendable. Scalar arrays took the overwrite branch.

Fix

Treat any non-empty array as extendable when the next key is an index ([n]), matching object-append behavior and issue #267.

Test plan

• go test -run TestSet .
• TestSetAppendsScalarToExistingArrayByIndex
• updated supplemental coverage expectation to append (not overwrite)

Fixes #267

[buger]

Closing as already-fixed — thank you for the fix, @Solaris-star.

The same root cause (Set on an array-index path beyond the current length silently overwriting instead of appending) was fixed in #283 as SYS-REQ-110, with three additional dimensions your PR didn't cover:

1. Scalar arrays — your fix handled the array-of-objects case (data[startOffset+1+nextToken(...)] == '{'); fix(parser): guard Delete leading-comma panic + empty-string key path components; L3 strict proof review #283 also fixes scalar arrays ([1,2,3]), which fell through to the overwrite path. Set({"a":[1,2,3]}, 99, "a", "[9]") now correctly returns {"a":[1,2,3,99]} instead of {"a":[99]} (silent data loss).
2. Cross-type auto-coerce — Set({}, 9, "[5]") previously produced malformed JSON ({[9]}); fix(parser): guard Delete leading-comma panic + empty-string key path components; L3 strict proof review #283 auto-coerces the container type so the output is always valid JSON ([9]). See KI-3 / DEFECT-260726-MFPA.
3. Top-level arrays — Set([1,2,3], 9, "[5]") still returns KeyPathNotFoundError; tracked as KI-4 (medium) pending a design call on whether top-level append is in scope.

Regression coverage: set_spec_test.go:TestSetAutoCoerce_KI3 + TestSetArrayIndexUnderObjectMalformedJSON_KI3 + the structure-aware fuzzer's output-validity gate. Filed as DEFECT-260727-WWWY.

Appreciate the contribution — the original issue (#267) was the right call, and your fix pointed at the correct code region.