perf: optimize charAt comparisons — 1,833x faster character parsing

[cs01]

Summary

• charAt(i) === "x" pattern: bypasses heap allocation entirely — emits direct getelementptr + load i8 + icmp instead of GC_malloc_atomic(2) + memcpy + strcmp
• Cached strlen via C bridge: cs_cached_strlen() caches the last (pointer, length) pair so repeated strlen() calls on the same string (the hot path in any parser) return in O(1) instead of O(n)
• Single-char string comparison: str === "x" uses byte compare instead of strcmp

Benchmark (JSONC parser, 10K keys / 237KB)

Before	After	Speedup
6.41s	0.01s	641x

On 2.5MB input: 0.03s (was 55s+ before — 1,833x faster)

Competitive with C (0.005s) and Go (0.01s) on the same workload.

Test plan

• New test fixture: string-charat-compare.ts — verifies charAt comparison correctness (in-bounds, OOB, negative index, equality/inequality)
• All existing string tests pass (charAt, charCodeAt, string-index)
• Self-hosting Stage 0 + Stage 1 pass
• JSONC benchmark verified

🤖 Generated with Claude Code