cua_s1: load and validate native vision checkpoints on CPU

[Levius-Fubuki]

Purpose

Add the native CPU VisionCheckpoint loader for pinned Qwen3.5-4B weights and the Cua-S1 multimodal adapter. Validate configuration, tensor inventories, dtypes, shapes, offsets, shard mappings and path containment; expose borrowed views over owned memory maps. Keep 297 BF16 base tensors and 100 FP32 LoRA tensors separate.

The diff contains the loader, its public module export and the required Serde derive dependency. Encoder execution is provided separately by #64.

Validation

After cleanup: workspace formatting, strict all-target Clippy, workspace tests and the locked release build passed. 19 tests passed; 3 existing GPU tests skipped. Runtime source was compared against the previous head: only new trailing test modules were removed; production code is unchanged. Existing baseline tests remain.

Feature-specific tests, examples, documentation and validation assets were removed from this PR diff to keep it focused on core implementation. They remain in the pre-cleanup commit and a local archive.

Before cleanup, all 397 visual tensor payload hashes matched an independent Python calculation; 13 loader integration tests passed. Those historical checks used the pre-cleanup test harness.

Self-review

Cleanup reviewed for unchanged runtime code, retained licensing and valid build targets. Draft status is retained for contributor/maintainer review.

• I have reviewed the full diff and addressed the issues I found.
• I have checked that the change follows the project's architecture and stays focused on the stated purpose.
• I have run the checks appropriate to this change and reported commands, results, and anything I could not verify above.
• I have checked that the PR description, documentation, and any accuracy or performance claims match the implementation and available evidence.