[lapack][rocSOLVER] Implement LU batch routines

[zjin-lcf]

Summary

• implement strided and grouped getrf_batch, getrs_batch, and getri_batch for buffer and USM APIs
• use rocSOLVER 64-bit batched getrf/getrs entry points on rocSOLVER 3.26+, with legacy fallbacks for older releases
• retain checked 32-bit pivot conversion for getri, which has no 64-bit rocSOLVER entry point
• make USM temporary-storage lifetimes explicit and add the corresponding scratchpad queries

Extracted from #767 for focused review. Depends on #778 for shared batched info handling and version detection; until that PR merges, this branch intentionally includes its foundation commit.

Test plan

• clang-format 19.1 check
• git diff --check
• rocSOLVER backend and RT/CT test binaries build for gfx90a
• focused getrf_batch/getrs_batch runtime tests on MI210 with ONEAPI_DEVICE_SELECTOR=hip:*: 40 passed
• local getri_batch execution is blocked by an assertion in the custom DPC++ runtime's device-kernel registration before the rocSOLVER call; CI should exercise this path with the supported toolchain

The legacy pre-3.26 fallback is retained for compatibility; the local ROCm 7.1 installation exercises the 64-bit getrf/getrs paths.