Update benchmark README with GPU performance results

[krystophny]

Problem

The benchmark README (bench/spline_many/README.md) references historical single runs but lacks current, verified performance numbers.

Current GPU Performance (RTX 5060 Ti, nvfortran 25.11)

Spline	CPU (pts/s)	GPU OpenACC (pts/s)	Speedup
1D	48.7M	807M	~16×
2D	8.4M	539M	~64×
3D	1.4M	102M	~73×

Detailed Results

1D Benchmark (order=5, num_points=2048, nq=8, npts=2M, niter=20, periodic=T):

• CPU: 48.7M pts/s
• OpenACC GPU: 807M pts/s

2D Benchmark (order=[5,5], num_points=[256,256], nq=8, npts=500K, niter=10, periodic=[T,T]):

• CPU: 8.4M pts/s
• OpenACC GPU: 539M pts/s

3D Benchmark (order=[5,5,5], num_points=[48,32,32], nq=8, npts=200K, niter=6, periodic=[T,T,T]):

• CPU: 1.4M pts/s
• OpenACC GPU: 102M pts/s

Proposed Changes

1. Add a "Performance Results" section with tabular data
2. Document test environment (GPU model, driver version, compiler version)
3. Note that nvfortran is the recommended compiler for GPU acceleration
4. Document GCC16 GPU status (currently has runtime issues - see Fix GCC16 OpenACC GPU offloading runtime error #209)