records: reconcile merged ROCm Q8_K state and remove stale operation totals

[VikashLoomba]

Row: BACKEND-ROCM

Problem

Two public/row projections are stale after recent ROCm merges:

• .agents/specs/rocm-q8k-cooperative-quantizer.md still says the implementation is not upstream. Pull request perf(BACKEND-ROCM): cooperate across Q8_K activation blocks #2472 merged as 9f96b74465441ebbee3651f4b316cdb0bf183715 on 2026-09-02. GitHub briefly marked QuantizeQ8KK is thread-per-superblock: 95.5us/call against llama.cpp's 1.6us, and 41% of the ROCm decode gap #1876 closed because a negated phrase in the merge message contained an adjacent closing keyword and issue number; localai-org-maint-bot reopened it at 2026-09-02T19:27:25Z. Issue QuantizeQ8KK is thread-per-superblock: 95.5us/call against llama.cpp's 1.6us, and 41% of the ROCm decode gap #1876 remains open and owns the pending gfx1200/gfx1201 arm.
• The Q8_K spec also says QuantizeQ8KK is thread-per-superblock: 95.5us/call against llama.cpp's 1.6us, and 41% of the ROCm decode gap #1876 lacks a canonical Row: line, but its body now begins with Row: BACKEND-ROCM.
• docs/FEATURES.md stores two hard-coded ROCm operation counts of 52. The wrapped-call-safe, device-specific command documented by docs/ROCM.md's op recount command is line-based, so it misses every multi-line RegisterOp( call -- and the count beside it is stale by nine #2573 currently finds another value, and any stored count will drift again when rocm_ops.hip changes.

Issue #2598 was opened while #1876 was temporarily closed. It is now closed as a duplicate because reopened #1876 owns the exact external-hardware scope.

Scope

• Reconcile the Q8_K spec's ## Now, issue-ownership text, outcome, and ## Owed section with merged perf(BACKEND-ROCM): cooperate across Q8_K activation blocks #2472, open/reopened QuantizeQ8KK is thread-per-superblock: 95.5us/call against llama.cpp's 1.6us, and 41% of the ROCm decode gap #1876, and its canonical row ownership.
• Remove hard-coded ROCm operation totals from both docs/FEATURES.md cells. Describe the native surface and link readers to docs/ROCM.md for the wrapped-call-safe recount command instead of storing a new count.
• Preserve the gfx1100-only acceptance boundary and the legacy unset default for gfx1200/gfx1201.
• Change no product code, tests, matrix lifecycle state, or unrelated documentation.

Gate

• The device-specific scan must execute successfully on the implementation base.
• No hard-coded ROCm registered-operation total may remain in docs/FEATURES.md.
• Repository prose, link, record, and full preflight gates must pass or retain only precisely classified inherited/argument-dependent results.

This is a record-only repair. It closes when the corrected projections merge.