fix: build CUDA kernels as multi-arch fatbin with PTX fallback

[0ax1]

No description provided.

[codspeed-hq]

Merging this PR will improve performance by 24.49%

⚠️ Unknown Walltime execution environment detected

Using the Walltime instrument on standard Hosted Runners will lead to inconsistent data.

For the most accurate results, we recommend using CodSpeed Macro Runners: bare-metal machines fine-tuned for performance measurement consistency.

⚡ 4 improved benchmarks
✅ 1233 untouched benchmarks

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
⚡	Simulation	chunked_varbinview_opt_canonical_into[(1000, 10)]	225.1 µs	187.9 µs	+19.79%
⚡	Simulation	fast_eq_out_of_range[4, 65536]	246.3 µs	188.5 µs	+30.65%
⚡	Simulation	fast_lt_out_of_range[16, 65536]	306.6 µs	248.8 µs	+23.24%
⚡	Simulation	fast_eq_out_of_range[16, 65536]	291.3 µs	233.9 µs	+24.53%

Tip

Curious why this is faster? Comment @codspeedbot explain why this is faster on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing fix-cuda-ptx-gpu-invalidation (b890192) with develop (a8fb30e)}