feat(clean): Remove OS state by default

[aaljewarane]

Summary

ec2-macos-init clean now also invokes ec2-macos-utils system cleanup-state by default, with subprocess STDIO passed through to the calling shell. A new -init-state-only flag limits cleanup to init's own state.

This fixes the recurring issue where AMIs built after clean still carried macOS's cached network interface configuration (NetworkInterfaces.plist), delaying or breaking networking on instances launched from those images.

Behavior:

• OS cleanup always runs (with or without init history), so a cleaned system is always ready for imaging. Absent instance history is treated as nothing to remove instead of a fatal error, and a history removal failure is logged, OS cleanup still runs, and clean exits non-zero so a partially cleaned system is never reported as success.
• If ec2-macos-utils is missing or fails, clean exits non-zero with a pointer to -init-state-only; a half-cleaned image is the bug being fixed.
• ec2-macos-utils is resolved from its install path (/usr/local/bin/ec2-macos-utils) first, falling back to PATH, since sudo's default PATH excludes /usr/local/bin for the documented sudo ec2-macos-init clean usage.

Dependency

Requires aws/ec2-macos-utils#41 to merge and release first.

Testing

• Builds darwin/amd64 + arm64; go vet clean; existing tests pass.
• Validated on mac1.metal (x86_64, macOS 15.7.7) and mac2.metal (arm64, macOS 26.3.1): removal, idempotence, plist regeneration on reboot, STDIO pass-through, -init-state-only opt-out, non-zero exit when utils is missing, history-failure ordering (OS cleanup still runs, then non-zero exit), install-path-preferred resolution, and PATH fallback.
• End to end on mac1: clean -> AMI -> instance launched from the derived image booted with healthy networking and a clean first boot.

[aaljewarane]

Note: Lint failures are pre-existing on master.

[aaljewarane]

Validation on EC2 Mac hardware

Validated ec2-macos-init clean on:

Family	Arch	macOS
mac1.metal	x86_64	15.7.7
mac2.metal	arm64	26.3.1

New init built from this branch (b63e025). "Old utils" is a real build of ec2-macos-utils origin/main (f5848d0, predates system cleanup-state) — confirms the --help probe against genuine cobra behavior, not a mock. Identical results on both families.

Backwards-compat / three-case handling

Case	Setup	Expected	Result
A	new utils installed	history + plist removed via delegation, STDIO passed through, exit 0	PASS
B	old utils (no command)	probe → warn + continue, exit 0, plist retained	PASS
C	no utils installed	warn + continue, exit 0, plist retained	PASS
D	utils present, command errors	hard-fail exit 1 (history already removed first)	PASS

Original functionality + flag

Case	Expected	Result
default clean	removes current instance history	PASS
-init-state-only	removes history, skips OS state, plist retained, exit 0	PASS
-all (multiple dirs)	removes all instance history, exit 0	PASS
-all, history absent	"nothing to remove", exit 0	PASS

Sample (case B, old utils):

Skipping OS state cleanup: installed ec2-macos-utils does not support 'system cleanup-state'
Cached OS state (e.g. NetworkInterfaces.plist) was not removed and may affect instances launched from an image of this system; update ec2-macos-utils to enable this cleanup
Clean complete   # exit 0

Sample (case D, real error):

Removing OS state with ec2-macos-utils
Unable to remove OS state: ec2-macos-utils system cleanup-state failed: exit status 1
Clean incomplete: unable to remove OS state: ...   # exit 1

Note: to avoid reboots, cases after the first real removal used a placeholder at the plist path; the Lstat+remove path is identical, and the first dry-run/removal exercised the genuine boot-generated plist.

Further notes: Escape hatch is still -init-state-only if someone needs to bypass.