Keep Laya weights and workspace on CUDA - #48

Purpose

Upload Laya's 205 used tensors once and allocate 17 reusable workspace buffers. Validate shapes before allocation and release partial allocations on failure.

Part of #14. Depends on #21 and #47. Kept in draft until those merge.

The diff against main currently includes those dependencies. Review this increment: 190 core/configuration lines for residency and workspace.

Tests now live under the repository-root tests/ directory. Cargo target names and test coverage are unchanged.

Test Plan

cargo fmt --all --check
cargo clippy --workspace --locked --all-targets -- -D warnings
cargo test --workspace --locked
cargo build --workspace --release --locked

On H800, compare uploaded tensors with the Torch conversion oracle and check workspace capacity and reuse. Reproduction commands are in src/models/laya/README.md.

System1-Omni Version / Commit: 2f5ff4b; incremental base 9ef493e.

Test Result

fmt, strict Clippy, release build passed locally; 37 CPU tests passed and 7 checkpoint/GPU tests were skipped. The earlier H800 validation below uses the same Laya and CUDA source; GPU tests were not rerun for this test-directory update.

• All 205 uploaded tensor hashes matched the oracle.
• All 17 buffers passed two write/read rounds at three shapes, including 16×512.
• Weight allocations: 803.72 MiB. Largest workspace: 225.11 MiB. Both exclude CUDA and library overhead.

Rust CI, Docs build and benchmark harness tests passed for this update.

Self-review

Before marking this PR ready for review or requesting maintainer review, complete
the self-review checklist.
Keep the PR in draft while this work is incomplete.
For agent assistance, use the optional precheck-pr skill.

• 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.