fix(mlx): bump MLX pin to 81ba1c6a so mxfp8 block scales round up

[inureyes]

Why

The fp8 round-trip bound failed on every Metal host, M1 Ultra byte-identically to M5 Max, because the pinned MLX 9a795735 predates ml-explore/mlx#4353: Metal and CPU encoded the mxfp8 E8M0 block scale as round(log2(amax / 448)), so about half the blocks saturated their maxima, losing up to 1 - 2^-1/2. CUDA rounds up, which is why #1742 passed on GB10. The test was right, and requantize_block_fp8_weights, the only E8M0 quantize caller, was clipping vendor FP8 checkpoints on Metal. Widening the bound, as the issue proposed, was rejected.

What changed

• MLX pin 9a795735 to upstream main 81ba1c6a (99 commits). The seven overlays whose targets upstream touched are three-way merged and keep their deltas; the other 21 are unchanged upstream. metal/compiled.cpp also drops a leftover elem_to_loc_1<uint> that undid part of Fix compiled kernel correctness for negative-strided inputs ml-explore/mlx#3720, so its only delta is the mixed-dtype cast.
• Adaptations to what the new pin changes under the bridge:
  • gather_qmm gained global_scale ahead of sorted_indices ([Metal] global scale for qmm ml-explore/mlx#4458), so all 13 calls pass std::nullopt.
  • CUDA now needs cuSOLVER ([CUDA] Cholesky via cuSOLVER ml-explore/mlx#4208). link_cuda() names it, docs/installation.md lists it, and CI's link job now runs on pin and CUDA link-list changes, which is how this slipped past CI.
  • array::is_available() now throws and clears a failed launch's error ( Propagate CPU errors to events ml-explore/mlx#3742). The rejection sampler's drain now reads status, signal and error pointer instead, so a GPU fault no longer terminates the process from inside fused_sample or hides the error from its owner. A regression test aborts the process with the old drain and passes with the new one.
• The round-trip check moves into its own test, fp8_block_requantize_round_trip_stays_within_half_an_e4m3_step, with the same seed and shape. It states the derivation and asserts group_max <= 448 * scale per block.

Validation (M1 Ultra, macOS 27.0)

• fp8: the old pin fails the new test at block 0 (the maximum 4.8046875 scales to 615 and saturates). The new pin saturates 0 of 650 blocks, with a worst error of 0.0489 of the group max against 0.2928 before.
• Turbo launchers pass (max RMS 1.7263e-4 and 1.5259e-4).
• Teacher-forced logit traces, old pin vs new, on five checkpoints at widths 1, 8 and 256: 0 disagreements on decided positions. Over 4,096 positions on qwen3-30b-a3b, 1 of 1,720 decided positions differs, at the reference's rank 2, with perplexity -0.20%.
• Branches the short runs missed:
  • head-dim-512 decode past 1,024 keys: identical text.
  • GQA-8 decode past 8,192 keys: identical text, decode 55.6 to 60.8 tok/s.
  • head-dim-72 vision towers: image prefill 239 to 279 tok/s. Descriptions diverge into equally faithful text; decided answers are unchanged.
• Short-context throughput is within 0.6% on three checkpoints.
• Workspace gate: 10986 passed, 0 failed, 359 ignored. Clippy, fmt, both pin parsers and the cross-repo reference check are clean.
• CUDA was compiled and linked in CI but not run: OpenXLA feature link linked a --features cuda,xla-iree release binary on GB10 at the new pin, which covers the overlays and the cuSOLVER link. The green CUDA sm_70 compile check is a skip, because CUDA 13.0 cannot target sm_70.

The per-checkpoint numbers, method and derivations are in TECHNICAL_REPORTS/1772-mlx-pin-mxfp8-round-up-20260911.en.md.

Not validated, or reported and not fixed

• No CUDA execution. No M5 Max run: the generation-17 NAX paths changed upstream in this range are unmeasured.
• array_evaluated_bytes, the server's lookahead read, is another non-Result bridge function that now throws on a failed launch. It needs routing through the scheduler's step-failure path.
• The mixed-dtype cast in metal/compiled.cpp would cast a comparison's inputs to bool. This is latent, since no compiled function contains a comparison.

Closes #1769

[inureyes]

PR finalization

Test coverage

Added a regression test for the deferred drain's Failed branch: a_failed_stashed_launch_is_dropped_without_throwing_or_consuming_its_error in src/lib/mlxcel-core/src/sampling_rejection_tests.rs, backed by a new test-only cxx bridge hook (sampling_dispatch_stash_failed_launch_for_test plus two accessors) in mlx_cxx_bridge.cpp/.h and lib.rs, following the existing sampling_dispatch_drain_pending/sampling_dispatch_reset hook pattern. It builds a valid, signalled event carrying an error through MLX's public Event/Error API (no GPU JIT needed), stashes it into pending-verification slot 0, then asserts drain_pending_verification neither throws nor consumes the error and drops the slot. Proved by temporarily reverting the drain to call is_available(): the reverted build crashed the whole test process with libc++abi: terminating due to uncaught exception of type std::runtime_error: mlxcel test: synthetic launch failure, the exact ml-explore/mlx#3742 failure mode this PR works around, then restored. cargo test --profile test-fast --features metal,accelerate -p mlxcel-core --lib sampling_rejection_tests:: reports 28 passed with the fix restored. Commit a2db793.

Documentation completeness

Checked docs/supported-models.md (Qwen3.5/3.8 FP8 entry), docs/architecture.md, CONTRIBUTING.md, README.md, docs/environment-variables.md, docs/turbo-kv-cache.md, and docs/upstream/mlx-cuda-quantized-matmul-nondeterminism.md for stale MLX pin, CUDA link library, or fp8-accuracy claims. None name the old pin as a current-state fact or make an accuracy claim the pin bump invalidates; mentions are either mechanism descriptions that hold regardless of the pinned commit, or historical citations (CHANGELOG.md's 9a795735 entry, CONTRIBUTING.md's 2c46b953 CUDA-threading citation, the upstream bug report pinned to b7c3dd6d) that are provenance and correctly left alone, matching how CHANGELOG.md is only touched at release-prep commits rather than per-PR. docs/installation.md remains the one doc that needed the cuSOLVER update, already in place. No changes made beyond the test coverage commit.