perf: optimize cache for tensor conversion

[DockeyMint]

Summary

• Add TensorCache mechanism for thread-local caching of tensor conversions in pybind11_utils.h, keyed on data_ptr + shape + strides + dtype
• Skip expensive DeviceFromPybind11HandleImpl and full conversion path on cache hit
• Merge with existing InternedNames / CallMethodNoArgs optimization (PR perf: optimize generic tensor conversion #832)

Motivation

Tensor conversion from Python to native representation is a hot path. The cache avoids redundant metadata extraction and memory allocation when the same tensor is converted repeatedly, while preserving the low-level Python C API optimizations already on master.

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

CUDACXX=/usr/local/cuda/bin/nvcc CMAKE_BUILD_PARALLEL_LEVEL=32 \
SKBUILD_BUILD_DIR=build-pr-smoke-cuda13 \
python -m pip install '.[dev]' --no-build-isolation --no-deps \
  --config-settings=cmake.define.INFINI_RT_ROOT=/data/shared/baoming/workplace/new_ops/infinirt-prefix \
  --config-settings=cmake.define.INFINI_OPS_SMOKE_BUILD=ON \
  --config-settings=cmake.define.WITH_NVIDIA=ON \
  --config-settings=cmake.define.WITH_CPU=OFF

Successfully built InfiniOps
Successfully installed InfiniOps-0.1.0

CUDA_VISIBLE_DEVICES=4 python -m pytest tests -m smoke -q
120 passed, 28 skipped, 13867 deselected in 15.37s

Test Results on Supported Platforms

Platform	Affected	Build / Smoke Result	Full Result / Notes
NVIDIA	Yes	Build passed; 120 passed, 28 skipped,	Full suite not run
Iluvatar
MetaX
Cambricon
Moore
Ascend

Full `pytest` output (optional)

Benchmark / Performance Impact

• Hardware: NVIDIA A100‑SXM4‑80GB; no foreign GPU process was observed during the alternating A/B run.
• Workload: FP32 16x16 GEMM, explicit implementation index 0.

Notes for Reviewers

• The optimization remains framework‑neutral and preserves the existing duck‑typed tensor contract (data_ptr, shape, dtype, device, and stride).