add -v phase timing and gate .ll write behind -v/-g

[cs01]

Summary

• Adds a one-line phase breakdown (parse=Xms sema=Yms codegen=Zms ir-write=Wms llvm=Vms link=Ums total=Tms) printed at the end of chad build -v, so next time a compile is mysteriously slow you can see where the time goes instead of guessing.
• Stops writing the 38 MB .ll artifact to disk on every build. It's now only written under -v, -g, or ir subcommand — the cases where you'd actually want to read it. .build/ stays clean.

Why

chad build src/chad-native.ts takes ~27s. I suspected the .ll disk write was a chunk of that; turns out it's 10ms. The real answer (from the new phase line) is that LLVM -O2 optimization owns ~92% of the time. Good to know — and cheap to keep the instrumentation around so we don't have to re-derive it.

Example output under -v:

phases: parse=278ms sema=5ms codegen=1173ms ir-write=10ms llvm=24579ms link=573ms total=26621ms

Test plan

• npm run verify:quick passes (tsc + stage 0 + stage 1 + unit tests)
• chad build foo.ts (no flags) produces no .ll file and no phase line
• chad build foo.ts -v writes .ll and prints phase line
• chad build foo.ts -g writes .ll (debug path)
• CI green

[github-actions]

Benchmark Results (Linux x86-64)

Benchmark	C	ChadScript	Go	Node	Bun	Place
Binary Trees	1.540s	1.213s	2.726s	1.274s	1.048s	🥈
Cold Start	1.2ms	1.0ms	1.3ms	29.6ms	12.8ms	🥇
Fibonacci	0.908s	1.409s	1.753s	3.380s	2.056s	🥈
File I/O	0.084s	0.092s	0.087s	0.166s	0.180s	🥉
JSON Parse/Stringify	0.004s	0.005s	0.015s	0.015s	0.007s	🥈
Matrix Multiply	0.490s	0.809s	0.530s	0.437s	0.426s	#5
Monte Carlo Pi	0.448s	0.439s	0.457s	2.594s	6.483s	🥇
N-Body Simulation	1.753s	2.260s	2.286s	2.350s	3.317s	🥈
Quicksort	0.246s	0.283s	0.242s	0.297s	0.261s	#4
SQLite	0.323s	0.371s	—	0.396s	0.400s	🥈
Sieve of Eratosthenes	0.015s	0.027s	0.020s	0.042s	0.035s	🥉
String Manipulation	0.008s	0.045s	0.016s	0.039s	0.030s	#5

CLI Tool Benchmarks

Benchmark	ChadScript	grep	node	xxd	Place
Hex Dump	0.431s	—	0.993s	0.144s	🥈
Recursive Grep	0.026s	0.012s	0.108s	—	🥈