Add ELF output format support for NPU2 (AIE2P) devices

ELF is a newer, self-contained XRT binary format that replaces the
two-file xclbin+insts.bin workflow. It embeds instructions via PDI
and simplifies deployment. ELF is only supported on NPU2 (AIE2P/Strix).

NPU2 now defaults to ELF format. NPU1 (AIE2) continues to use xclbin
unchanged. The format can be overridden via AMD_TRITON_NPU_OUTPUT_FORMAT.

Key changes to driver.py:
- Add _get_output_format() for auto-detection and env var override
- Add _generate_elf_launcher() using xrt::elf, xrt::ext::kernel,
  xrt::ext::bo, and xrt::run APIs (no instruction binary needed)
- Add _extract_elf_kernel_name() to read kernel name from ELF config
- Modify compile_module() for conditional linker flags, aircc flags
  (--output-format elf --elf-name), caching (2 vs 3 files), and
  module loading
- Modify NPULauncher to dispatch to correct launcher generator

Tested: 14/15 examples pass on NPU2 with ELF (matvec fails in both
ELF and xclbin modes -- pre-existing issue).

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: erwei-xilinx