[doc-only] docs(core): finalize 1.1.0 release notes

[Andy-Jost]

Summary

Finalizes the cuda.core v1.1.0 release notes. Part of the v1.1.0 release checklist (#2291), item "Finalize the doc update, including release notes".

The notes were assembled by walking every cuda_core/ squash-merged commit since v1.0.1 and grouping the user-facing changes into Highlights / New features / Fixes and enhancements / Documentation / Deprecated APIs, following the structure of the 1.0.0 notes. Each entry links to its originating PR (and issue, where one was closed).

Changes

• Highlights: .pyi type-stub support, the new cuda.core.texture module, richer managed-memory support (ManagedBuffer advice API + batched range ops), and CUDA 13.3 toolkit support.
• New features: Host managed-memory locations, ManagedBuffer and its property-style advice API, batched managed-memory range ops in cuda.core.utils, device-specific NVLink enumeration, graph.GraphBuilder.graph_definition, the cuda.core.texture module, .pyi stubs, path-like inputs for ObjectCode/Program, and a Buffer.size accessor.
• Fixes and enhancements (renamed from "Bug fixes" to match the 1.0.0 notes): WSL get_process_name fixes, NVLink link-number validation, IPC import hardening, ManagedBuffer.accessed_by torn-state fix, graph node attachment lifetimes, graph user-object destructor shutdown-race hardening, free-threading correctness fixes, FIPS-safe compile cache keys, memory-pool driver-error preservation, correct BufferError on DLPack export, Buffer/MemoryResource __eq__ corrections, early GPU-UUID validation on checkpoint restore, and the PyTorch tensor-bridge upper-bound bump.
• Documentation: IPC buffer pickle trust-boundary note (Buffer.__reduce__).
• Deprecated APIs: system.NvlinkInfo.max_links.

Test Coverage

Docs-only change (title marked [doc-only] to skip the wheel/test matrix). pre-commit run passes on the notes file, including the RST hooks and lychee link validation over all the new PR/issue URLs.

Related Work

Release checklist: #2291
Companion docs PR: #2314 (version switcher)

[leofang]

#2309 is merged

[Andy-Jost]

@mdboom please let me know if there's anything else from your work you'd like added to the highlights section

[mdboom]

Just a disagreement about what is a highlight, but other than that, looks good.

[Andy-Jost]

@mdboom I made the adjustments you suggested. I also added entries for the new texture module and changes to graph-attached objects.

[leofang]

Reviewing

[leofang]

Thanks for driving this! I walked the 77 squash-merged commits under cuda_core/ since v1.0.1 (342b3199) and found several user-facing PRs that should be captured. Suggestions below, grouped following the 1.0.0-notes structure.

Highlights — one addition:

• CUDA 13.3 toolkit support (#2139) — append as a trailing bullet.

New features — two additions:

• ObjectCode / Program accept path-like inputs (#2123).
• Buffer.size accessor exposed to Python (#2068, closes #2049).

Bug fixes — additions + one link backfill:

• The existing "WSL / UnicodeDecodeError" bullet is #2118 — please backfill the PR link.
• Graph user-object destructor hardened against Python-shutdown races (#2074).
• Free-threading correctness: buffer/MR threading fixes (#2162), critical-section guards on shared accessors (#2215), atomic flag for buffer memory-attribute init (#2216).
• Program.compile cache keys are now FIPS-safe (#2087).
• Memory pool driver errors are preserved instead of masked by OOM handling (#2084).
• check_has_dlpack now raises BufferError (the intended exception) instead of RuntimeError (#2160).
• Buffer / MemoryResource __eq__ implementations corrected (#2067, closes #2050).
• Checkpoint restore validates GPU UUID inputs early (#2086).
• PyTorch tensor-bridge upper bound bumped to 2.12 (#2099).

Documentation — worth its own bullet:

• IPC buffer pickle trust-boundary documentation (#2225). Security-relevant note on Buffer.__reduce__; multi-process IPC users should see it.

Style nits:

• 1.0.0 notes call this section "Fixes and enhancements"; this draft uses "Bug fixes". Fine either way, just flagging the divergence.
• The existing .pyi stubs and NVLink get_nvlink validation bullets lack PR links; for consistency with the New-features entries, please add #2061 and #2192 respectively.

Requesting changes to bundle these additions before landing.

[Andy-Jost]

@leofang thanks for the sweep. Everything you identified is in place now.

[leofang]

(I applied #2319 to other API ref sections and re-ordered the API refs so that the compiler toolchain shows up earlier. It does not make sense to order it after texture/object.)

[github-actions]

Doc Preview CI
Preview removed because the pull request was closed or merged.