[FEA] cuda.core: add remaining new NVRTC options to ProgramOptions (CUDA 13.1+)

[leofang]

Summary

ProgramOptions kept pace with NVRTC 12.9 (ofast_compile, frandom_seed, no_cache are all
present), but not with the 13.x option additions:

Option	Documented since	Note
--warning-as-error (-Werror)	13.1	promote warnings to errors
--Wreorder	13.1	member-initialization-order warning
--enable-tile, --tile-only, --simt-only, --default-tile, --diagnose-implicit-tile-var	13.3	tile compilation — coordinate with #1322 / #1434 rather than adding blind
--use-bundled-headers=<dir>	13.3	scoped in #2363

New options must be version-gated with actionable errors per the mechanism tracked in #337.

Design sketch (draft — needs design-meeting review)

Important

Starting point only, not a settled design — review in the cuda.core design meeting.

• warning_as_error: bool | str | list[str] (bare -Werror vs. -Werror=<kinds> — check the
  exact accepted forms against the 13.x NVRTC docs during implementation).
• wreorder: bool.
• Tile flags: hold until the Program-side Tile IR design (cuda.core.Program: Support Tile IR #1322) settles, then add as one
  coherent group.

Open questions for the meeting:

1. Reject unsupported options at ProgramOptions construction or at compile() (per Add descriptive exception handling to linker and program options which are not supported on a CTK version basis + handle in tests #337)?
2. Are the tile flags meaningful without the Tile IR output path, or strictly coupled to cuda.core.Program: Support Tile IR #1322?

References

• NVRTC docs (supported compile options): https://docs.nvidia.com/cuda/nvrtc/index.html
• Found during the CUDA 12.8 → 13.3 bindings vs. cuda.core gap sweep (2026-07-14)

-- Leo's bot