feat(ascend): add reshape_and_cache_flash provider

[baominghelly]

Summary

• Add the Ascend reshape_and_cache_flash provider in src/native/ascend/ops/reshape_and_cache_flash/kernel.h.
• Use ACLNN ScatterPaKvCache for contiguous non-float32 inputs and an ACLNN ScatterNdUpdate fallback for float32 or non-contiguous layouts.
• Preserve negative slot mappings by skipping the corresponding cache updates.

Motivation

Ascend model inference needs native KV-cache population for the existing reshape_and_cache_flash interface. Without this provider, paged-attention cache updates cannot use the InfiniOps slot-0 path on Ascend.

Type of Change

• feat — new feature / new operator / new platform
• fix — bug fix
• perf — performance improvement (no behavioral change)
• refactor — code restructuring without behavior change
• test — adding or fixing tests only
• docs — documentation only
• build / ci — build system or CI configuration
• chore — tooling, formatting, or other non-code changes
• Breaking change (requires a ! in the Conventional Commits prefix or a BREAKING CHANGE: footer)

Platforms Affected

• CPU (WITH_CPU)
• NVIDIA (WITH_NVIDIA)
• Iluvatar (WITH_ILUVATAR)
• MetaX (WITH_METAX)
• Cambricon (WITH_CAMBRICON)
• Moore (WITH_MOORE)
• Ascend (WITH_ASCEND)
• PyTorch C++ bindings (WITH_TORCH)
• Build system / CMake / CI
• Python bindings / user-facing API

Smoke Test Result

bash /workspace/ascend-slot0/configure-focused-build.sh \
  /workspace/ascend-slot0/reshape-and-cache-flash \
  /workspace/ascend-slot0/build-reshape-and-cache-flash \
  /workspace/ascend-slot0/ops-reshape-and-cache-flash.json
cmake --build /workspace/ascend-slot0/build-reshape-and-cache-flash --parallel 32
bash /workspace/ascend-slot0/basic-ops-20260901/scripts/run-focused-test.sh \
  /workspace/ascend-slot0/reshape-and-cache-flash \
  /workspace/ascend-slot0/build-reshape-and-cache-flash \
  /workspace/ascend-slot0/python-reshape-and-cache-flash \
  tests/test_reshape_and_cache_flash.py

[5/5] Linking CXX shared module src/ops.cpython-311-aarch64-linux-gnu.so
24 passed, 14 warnings in 1.43s

Test Results on Supported Platforms

Platform	Affected	Build / Smoke Result	Full Result / Notes
NVIDIA	No	N/A - not affected	N/A - not affected
Iluvatar	No	N/A - not affected	N/A - not affected
MetaX	No	N/A - not affected	N/A - not affected
Cambricon	No	N/A - not affected	N/A - not affected
Moore	No	N/A - not affected	N/A - not affected
Ascend	Yes	Focused build passed on Ascend 910C with CANN 9.0.0	Complete tests/test_reshape_and_cache_flash.py: 24 passed

Full `pytest` output (optional)

collected 24 items
24 passed, 14 warnings in 1.43s

Benchmark / Performance Impact

N/A. This PR adds functional provider coverage and does not claim a performance improvement.

Notes for Reviewers

• The fast path requires contiguous key/value/cache tensors and excludes float32.
• The fallback copies slot mappings to host, skips negative slots, constructs block/offset indices, and applies per-token scatter updates.
• Validation covered float16, bfloat16, and float32, both NHD/HND layouts, and contiguous/non-contiguous cases in the complete operator test file.
• The repository-wide test suite was not run.