add llvm.assume bounds hints, full fast-math flags on all fp ops

[cs01]

Summary

• @llvm.assume after bounds checks — after a successful bounds check, emit call void @llvm.assume(i1 %in_bounds) to explicitly tell LLVM's optimizer that the index is valid. Zero runtime cost (assume generates no machine code). Enables SCEV to potentially eliminate redundant bounds checks in loops with monotonic access patterns.
• Full fast-math flags on all floating-point operations — add nnan ninf to the existing nsz arcp contract reassoc afn flags on fadd, fsub, fmul, fdiv, frem, fneg. This tells LLVM it can assume no NaN/Infinity values, enabling more aggressive loop transforms and FP optimizations.
• IR builder consistency — the emitFAdd/emitFSub/emitFMul/emitFDiv/emitFRem/emitFNeg builder functions now emit the same fast-math flags as the binary operator codegen path.

These are foundational optimizations that give LLVM's optimizer more information to work with. They don't produce measurable speedups on current benchmarks alone, but they're prerequisites for future optimizations (loop bounds check elimination, auto-vectorization) and ensure consistent optimization across all FP code paths.

Test plan

• npm run verify:quick passes (tests + self-hosting Stage 1)
• All existing tests pass unchanged (fast-math and assume are purely optimizer hints)

[github-actions]

Benchmark Results (Linux x86-64)

Benchmark	C	ChadScript	Go	Node	Bun	Place
Binary Trees	1.390s	1.239s	2.710s	1.201s	0.934s	🥉
Cold Start	0.8ms	0.8ms	1.2ms	24.1ms	9.3ms	🥇
Fibonacci	0.818s	1.469s	1.561s	3.186s	1.943s	🥈
File I/O	0.119s	0.090s	0.088s	0.203s	0.174s	🥈
JSON Parse/Stringify	0.004s	0.005s	0.018s	0.015s	0.007s	🥈
Matrix Multiply	0.457s	0.699s	0.634s	0.365s	0.329s	#5
Monte Carlo Pi	0.389s	0.410s	0.404s	2.247s	6.028s	🥉
N-Body Simulation	1.669s	2.119s	2.202s	2.386s	3.260s	🥈
Quicksort	0.214s	0.245s	0.213s	0.262s	0.226s	#4
SQLite	0.350s	0.400s	—	0.431s	0.401s	🥈
Sieve of Eratosthenes	0.015s	0.027s	0.020s	0.039s	0.035s	🥉
String Manipulation	0.008s	0.045s	0.016s	0.037s	0.027s	#5

CLI Tool Benchmarks

Benchmark	ChadScript	grep	node	xxd	Place
Hex Dump	0.430s	—	1.018s	0.128s	🥈
Recursive Grep	0.019s	0.009s	0.094s	—	🥈