Migrate logging from internal/logger to log/slog for amd-ctk

[shiv-tyagi]

Replace the custom logger.Log (backed by standard log.Logger) with Go's structured logging (log/slog) across the project.

• Configure slog in cmd/amd-ctk with --debug flag for level control (defaults to Info on stderr)
• Configure slog in cmd/container-runtime via internal/logger.Init for file-based logging at Debug level to preserve the previous behavior where all log lines were captured
• Update internal/logger to set slog as the default logger instead of exposing a log.Logger instance
• Convert all logger.Log.Printf calls to appropriate slog levels (Info, Debug, Warn, Error) with structured key-value pairs
• Remove logger.Log references from all internal packages and tests

[shiv-tyagi]

Test Checklist

GPU Tracker — Enable/Disable/Reset

• amd-ctk gpu-tracker init — succeeds silently (no output, exit 0)
• amd-ctk gpu-tracker enable — prints "GPU Tracker has been enabled"
• amd-ctk gpu-tracker enable (again) — prints "GPU Tracker is already enabled"
• amd-ctk gpu-tracker disable — prints "GPU Tracker has been disabled"
• amd-ctk gpu-tracker disable (again) — prints "GPU Tracker is already disabled"
• amd-ctk gpu-tracker reset (was disabled) — prints "GPU Tracker has been reset" (no restart recommendation)
• amd-ctk gpu-tracker enable then amd-ctk gpu-tracker reset — prints "GPU Tracker has been reset" followed by restart recommendation
• All above as non-root — prints "Permission denied: Not running as root"

GPU Tracker — Status

• amd-ctk gpu-tracker status (disabled) — prints "GPU Tracker is disabled"
• amd-ctk gpu-tracker status (enabled, GPUs present) — prints table with correct GPU Id, UUID, Accessibility, Container Ids
• amd-ctk gpu-tracker status (GPU info mismatch, e.g. GPU removed after enable) — prints error: "GPU info mismatch: please reset GPU Tracker"

GPU Tracker — Exclusive/Shared

• amd-ctk gpu-tracker 0 exclusive — prints "GPUs [0] have been made exclusive"
• amd-ctk gpu-tracker 0,1 exclusive (GPU 0 used by 2+ containers) — prints "GPUs [0] have not been made exclusive because more than one container is currently using it"
• amd-ctk gpu-tracker 0,99 exclusive — prints success for 0, prints "Ignoring [99] GPUs as they are invalid"
• amd-ctk gpu-tracker 0 shared — prints "GPUs [0] have been made shared"
• amd-ctk gpu-tracker 0,1 exclusive (tracker disabled) — prints "GPU Tracker is disabled" (no operation performed)
• amd-ctk gpu-tracker 0,1 exclusive (GPU info mismatch) — prints error: "GPU info mismatch: please reset GPU Tracker"

GPU Tracker — Release

• amd-ctk gpu-tracker release <container_id> — succeeds silently (exit 0)
• amd-ctk gpu-tracker release <container_id> (tracker file missing) — succeeds silently (tracker not initialized, nothing to release)
• amd-ctk gpu-tracker release <container_id> (permission denied on tracker file) — prints error with container ID context

GPU Tracker — Error Scenarios

• Delete /var/log/gpu-tracker.json and run amd-ctk gpu-tracker status — auto-initializes and shows status (or "disabled")
• Make /var/log/gpu-tracker.json immutable and run amd-ctk gpu-tracker enable — prints clean error (no "Error: Error:" nesting)
• Corrupt /var/log/gpu-tracker.json with invalid JSON and run amd-ctk gpu-tracker status — prints clean error: "decoding GPU tracker JSON: ..."
• Hold the lock file (flock /var/log/gpu-tracker.lock sleep 30 &) and run amd-ctk gpu-tracker enable — times out after 10s with: "acquiring lock: timeout exceeded"

GPU Tracker — JSON Backward Compatibility

• With an existing /var/log/gpu-tracker.json (from old version with "accessibility": 0), run amd-ctk gpu-tracker status — deserializes correctly, shows "Shared"
• Same file with "accessibility": 1 — shows "Exclusive"
• Run amd-ctk gpu-tracker init — overwrites with new file, still uses int format ("accessibility": 0)

Container Runtime — GPU Allocation

• docker run --runtime=amd -e AMD_VISIBLE_DEVICES=all <image> rocm-smi — GPUs visible inside container
• docker run --runtime=amd -e AMD_VISIBLE_DEVICES=0 <image> rocm-smi — only GPU 0 visible
• docker run --runtime=amd <image> echo hello (no AMD_VISIBLE_DEVICES) — container starts without GPUs, no errors
• Check /var/log/amd-container-runtime.log — contains structured slog entries: "Creating ROCm container runtime", "Launching container"
• Verify no output on container's stdout from the runtime itself

Container Runtime — Exclusive GPU Enforcement

• Enable tracker, make GPU 0 exclusive
• Start container 1 with GPU 0 — succeeds, logs "GPUs allocated gpus=[0]"
• Start container 2 with GPU 0 — fails with error: "GPUs [0] are exclusive and already in use"
• Stop container 1 — poststop hook releases GPU 0
• Start container 2 with GPU 0 — now succeeds

Container Runtime — Log File

• As root: verify log written to /var/log/amd-container-runtime.log
• Verify log format is structured: time=... level=... msg="amd-container-runtime ..." key=value
• Verify Debug-level messages appear in log file (e.g. "Running runc", "Parsing topology node file")

CDI — Generate/Validate

• amd-ctk cdi generate — prints "Generated CDI spec: /etc/cdi/amd.json" (or custom path)
• amd-ctk cdi generate --output /tmp/amd.json — generates at custom path
• amd-ctk cdi validate — prints "Validating CDI spec: ..." then "CDI spec is valid"
• Modify the generated spec and run amd-ctk cdi validate — prints "CDI spec is invalid\nPlease regenerate CDI spec"
• amd-ctk cdi validate with missing spec file — prints error: "validating CDI spec: opening CDI spec file ...: no such file or directory"
• amd-ctk cdi validate with corrupt JSON — prints error with decode details

CLI — Debug Flag

• amd-ctk gpu-tracker status — no debug output on stderr
• amd-ctk --debug gpu-tracker status — debug-level slog output visible on stderr
• amd-ctk -d gpu-tracker status — same as above (short flag)

Signal Handling

• Start amd-ctk gpu-tracker enable and send SIGINT (Ctrl+C) during execution — process exits, lock file released (verify by immediately running another command)
• Start amd-ctk gpu-tracker enable and send SIGKILL — lock file released by kernel (verify by immediately running another command)

Unit Tests

• go test ./internal/gpu-tracker/ — passes
• go test ./internal/cdi/ — passes
• go test ./internal/amdgpu/ — passes
• go test ./internal/oci/ — passes
• go build ./cmd/... — builds without errors

I have performed these tests on a GPU machine and carefully verified all outputs.