perf: 2.5x faster string split on single-char separators

[cs01]

User-visible impact

"text".split("\n"), .split(","), .split(" ") — the most common split patterns — are now roughly 2.5x faster. Any ChadScript program that parses line-based or delimited data (log readers, CSV loaders, tokenizers, grep-style tools) gets the speedup for free.

Measured on the string search benchmark (recursively grep console.log in src/)

binary	before	after
chad	~21 ms	~8.5 ms
C baseline	~7 ms	~7 ms

Chad now runs within ~20% of a hand-written C implementation of the same benchmark. Instruction count drops from 267M to 101M; cycles from 91M to 37M.

Root cause

cs_str_split (the C bridge backing String.prototype.split) walked the source buffer one byte at a time and called memcmp(src + pos, sep, 1) on every byte when the separator was a single character. That's a PLT-stub call per byte, and memcmp with n=1 never reaches its SIMD fast path. For a typical 2-3 KB source file that's thousands of stub calls per split — profiling the old benchmark showed ~80% of total time in cs_str_split → _platform_memcmp.

Fix

Added a sep_len == 1 fast path that uses memchr instead. memchr is a libc SIMD primitive that scans whole cache lines at once (NEON on arm64, AVX2 on glibc x86_64), so finding the next separator in a long string costs ~1 cycle per 16 bytes instead of a full function call per byte. Both passes of cs_str_split (counting parts, then filling them) take the fast path.

Multi-char separators are unchanged.

Test plan

• npm run verify:quick — all tests + stage 1 self-hosting pass
• String search benchmark improves from 21 ms to ~8.5 ms
• String ops benchmark (also uses single-char split) runs clean at ~13 ms
• CI green

[github-actions]

Benchmark Results (Linux x86-64)

Benchmark	C	ChadScript	Go	Node	Place
Binary Trees	1.589s	1.300s	2.720s	1.157s	🥈
Cold Start	0.9ms	0.8ms	1.2ms	28.7ms	🥇
Fibonacci	0.816s	0.765s	1.393s	3.164s	🥇
File I/O	0.117s	0.093s	0.086s	0.205s	🥈
JSON Parse/Stringify	0.004s	0.005s	0.018s	0.015s	🥈
Matrix Multiply	0.445s	0.732s	0.614s	0.364s	#4
Monte Carlo Pi	0.389s	0.410s	0.405s	2.248s	🥉
N-Body Simulation	1.666s	2.122s	2.202s	2.400s	🥈
Quicksort	0.216s	0.247s	0.213s	0.262s	🥉
SQLite	0.355s	0.406s	—	0.424s	🥈
Sieve of Eratosthenes	0.015s	0.027s	0.020s	0.040s	🥉
String Manipulation	0.008s	0.037s	0.016s	0.037s	🥉

CLI Tool Benchmarks

Benchmark	ChadScript	grep	node	xxd	Place
Hex Dump	0.434s	—	0.995s	0.131s	🥈
Recursive Grep	0.019s	0.010s	0.099s	—	🥈