JSONEncoder: fixed _asDirectArrayEncodable function

[ChrisBenua]

In my recent PR, I optimized JSONDecoder/JSONEncoder. However, I introduced a small mistake in the _asDirectArrayEncodable function:
Instead of using the value parameter from the function arguments, I mistakenly referenced the array property from the __JSONEncoder class.
Since __JSONEncoder.array is of type JSONFuture.RefArray, all type equality checks (e.g. against [Int]) failed.
As a result, the intended optimization did not take effect.
This issue slipped past benchmarks because there are currently no tests for decoding large [Int] (or similar) arrays.

To address this, I’ve added an additional benchmark that encodes 100k Ints, which highlights the performance improvements.

array-encodeToJSON metrics

Time (total CPU): results within specified thresholds, fold down for details.

Time (total CPU) (ns) *	p0	p25	p50	p75	p90	p99	p100	Samples
new_as_direct_array_encodable_coders_v4	38	38	38	38	38	38	38	79
Current_run	3	3	3	3	3	3	3	985
Δ	-35	-35	-35	-35	-35	-35	-35	906
Improvement %	92	92	92	92	92	92	92	906

Throughput (# / s): results within specified thresholds, fold down for details.

Throughput (# / s) (K)	p0	p25	p50	p75	p90	p99	p100	Samples
new_as_direct_array_encodable_coders_v4	26	26	26	26	26	26	26	79
Current_run	330	329	329	329	327	322	298	985
Δ	304	303	303	303	301	296	272	906
Improvement %	1169	1165	1165	1165	1158	1138	1046	906

I’m happy to either include the new benchmark in this PR or submit it separately, depending on the maintainers’ preference. We can significantly reduce risks of introducing encoding large arrays performance degradations with separate benchmark.