fix(apijson): replace O(n^2) array encoding with O(n) bytes.Buffer

[ssicard]

Summary

newArrayTypeEncoder builds JSON arrays by calling sjson.SetRawBytes(json, "-1", item) in a loop. sjson re-parses and reallocates the entire accumulated buffer on each call, giving O(n^2) time and memory complexity. For a 30,000-element IP list update via Rules.Lists.Items.Update, this allocates ~6.2 GB and OOM-kills the process.

This replaces the sjson loop with a single-pass bytes.Buffer concatenation, restoring O(n). Output is byte-identical to the previous implementation.

Context

• Customer escalation: a SAP BTP controller syncing ~30,000 IPs is OOM-killed every ~6 minutes
• GitHub issue: O(n²) memory allocation when marshaling large slices via Marshal() #4315
• Customer PR with equivalent fix: fix(apijson): avoid O(n²) array marshaling for large slices #4316

Benchmarks (from #4316)

N	before B/op	after B/op	mem reduction	before ns/op	after ns/op	speedup
100	92 KB	8 KB	11x	138 us	24 us	5.9x
1,000	6.9 MB	66 KB	105x	7.1 ms	0.21 ms	33x
10,000	692 MB	845 KB	819x	437 ms	1.18 ms	369x
30,000	6.2 GB	2.0 MB	3,070x	4,091 ms	3.10 ms	1,320x

Tests

All existing internal/apijson tests pass. Added array_bench_test.go with correctness verification against encoding/json and benchmarks at N=100..30,000 as a regression guard.

Fixes #4315