Add logging knob to control console output

[erwei-xilinx]

Summary

• Replaces 11 print() calls in driver.py with Python's logging module, controlled by a new AMD_TRITON_NPU_DEBUG environment variable
• By default (unset or 0), only warnings are shown — no compilation messages clutter stdout during normal runs
• Set AMD_TRITON_NPU_DEBUG=1 to restore verbose output (cache hits, tiling script paths, compile-only mode details)
• Autotuning prints remain gated by knobs.autotuning.print and are unaffected

Closes #40

Test plan

• Run an example kernel without AMD_TRITON_NPU_DEBUG — verify no compilation prints on stdout
• Run with AMD_TRITON_NPU_DEBUG=1 — verify same output as before this change
• Run with AMD_TRITON_NPU_COMPILE_ONLY=1 + AMD_TRITON_NPU_DEBUG=1 — verify compile-only messages appear
• Run same kernel twice with AMD_TRITON_NPU_DEBUG=1 — verify cache hit message appears
• Verify autotuning prints still controlled by knobs.autotuning.print

🤖 Generated with Claude Code

[Copilot]

Pull request overview

This PR replaces several print() statements in the NPU backend driver with Python logging, introducing an AMD_TRITON_NPU_DEBUG environment variable intended to control whether compilation/cache-related messages appear on the console.

Changes:

• Added a module logger and an AMD_TRITON_NPU_DEBUG-controlled logging setup.
• Converted selected compilation/cache/tiling-script console prints to logger.info(...).
• Converted device-detection error prints to logger.warning(...).

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.