feat(linx): export memory results after test finisher

[jiale-wangOwO]

Summary

• add opt-in cross-model-dump, cross-model-address, and cross-model-size machine properties
• publish the requested RAM range atomically after a passing guest finisher shutdown
• reject invalid/oversized/out-of-RAM ranges and never publish a dump for timeout or non-finisher shutdown

This is the QEMU-side transport required by LinxISA/SuperScalarModel#31. It does not change default machine behavior and does not compare internal Tile identifiers.

Validation

• ninja -C build qemu-system-linx64
• same PTO ISA v0.2 ELF exports 512 bytes matching gfrun and independent golden data
• S32 8x8 TLOAD/TSTORE and TADD result SHA-256: c5e4b3cd5c84854090c78ee0df5c711a33617f7cb3c087dbe4a001f260c251a0
• missing/zero/oversized/out-of-RAM configurations reject at startup
• a two-second non-finisher timeout creates no result file

Dependency order

Merge this PR before the SuperScalarModel comparator PR. Advance the linx-isa QEMU gitlink only after this PR is merged, and point it at the upstream merge commit.

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.

[jiale-wangOwO]

Closing as superseded by #41.

PR #41 contains this PR's complete commit 9702fe53eaef2815c72dff53c248f9fd2106c7ad as an ancestor, including the opt-in finisher-time architectural memory dump used by the QEMU-gfrun-gfsim validation harness. It also carries the subsequent TEPL row-reduction, TCVT dtype, and canonical TMA profile fixes.

No functionality from this PR is being dropped. Please review and merge #41 as the single QEMU integration path. After #41 merges, the linx-isa gitlink should be advanced to its upstream merge commit, not either PR head.

Tracking: LinxISA/SuperScalarModel#31