fix(linx): align TEPL row reductions and TCVT profiles

[jiale-wangOwO]

Summary

• add finisher-time architectural result dumping used by the QEMU-gfrun parity harness
• publish row-reduction results with vector-shaped Tile metadata
• implement the PTO ISA v0.2 TCVT dtype contract: BSTART.TEPL is the source dtype and B.DATR is the destination dtype
• validate TCVT source profile before destination reservation or mutation

Validated profiles

The SuperScalarModel parity carrier executes the same ELF and initialized input in QEMU and gfrun, then compares both outputs independently with a manifest-generated golden result.

• 79 currently executable TEPL profiles across tile_smoke, tepl_smoke, and tepl_reduce
• TCVT: FP32 -> S32, RTZ, saturation disabled, 1x8 row-major
• row reductions and row/column broadcast profiles covered by tepl_reduce

Current QEMU canonical execution blockers remain explicitly unclaimed: TGATHER, TSCATTER, TTRANSPOSE, TINTERLEAVE, TDEINTERLEAVE, TPARTARGMAX, and TPARTARGMIN.

Verification

• ninja -C build qemu-system-linx64
• SuperScalarModel strict gfrun and gfsim builds
• SuperScalarModel CTest: 3/3 PASS
• cross-model regression: tile_smoke, tepl_smoke, tepl_reduce, and cube_smoke PASS
• cross-model harness unit tests: 9/9 PASS

Representative three-way hashes:

• tepl_smoke: c443b551e800ea699e15a2b8b02f8785e6b5d7d2cf24443f61480eccba3ab590
• tepl_reduce: 1bd25dbb990b5b0d3c8a0a7c03fe5223ab7ee814c1eed6231cc313dacd3a6a30

Tracking: LinxISA/SuperScalarModel#31
Related parity work: LinxISA/SuperScalarModel#76

[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]

PR #40 has been closed as superseded by this PR. Its complete memory-dump commit 9702fe53eaef2815c72dff53c248f9fd2106c7ad is retained as the first commit in this branch, so no validation transport functionality is lost.

This PR is now the single QEMU merge path for the architectural-result dump plus the TEPL/TCVT/TMA follow-up fixes. After merge, repin linx-isa to this PR's upstream merge commit.

Refs #40 and LinxISA/SuperScalarModel#31